The Role of CDS Trading in the Commercialization of New Lending Relationships

July 26th, 2018

The Role of CDS Trading in the Commercialization of New Lending Relationships

Abstract

We investigate how the development of the credit default swap (CDS) market affects lenders’ incentives to initiate new lending relationships. We predict that CDSs reduce the adverse selection that non-relationship lenders face when competing for loans, by allowing those lenders to hedge loan exposure and by the revelation of private information through CDS spreads. We find that, following CDS initiation on a borrower’s debt, non-relationship lead arrangers are more likely to originate its loans and non-relationship participants are more likely to join loan syndicates. We also show that lead arrangers that initiate lending relationships following CDS initiation focus more on commercial aspects of lending relationship. These lead arrangers are more likely to pursue new borrowers with high cross-selling potential, which are expected to generate substantial fee business. Further, non-relationship lenders have lower incentives for costly borrower monitoring, as reflected in weaker control rights and in the lower loan share they retain. Relative to relationship lenders, non-relationship lenders are likely to be more distant from borrowers, foreign, and less reputable once CDSs become available, emphasizing their lower monitoring efficiency.

1

1. Introduction

The development of the credit default swap (CDS) market is one of the most important

financial innovations of recent decades. CDSs allow lenders to hedge borrowers’ credit risk while

maintaining their lending relationships (e.g., Saretto and Tookes 2013). However, CDS hedging

reduces lenders’ incentives to monitor borrowers and may push borrowers into inefficient

bankruptcy (e.g., Hu and Black 2008; Ashcraft and Santos 2009; Bolton and Oehmke 2011;

Parlour and Winton 2013). Considering these positive and negative attributes, we examine how

CDSs influence the loan origination and participation decisions in loan syndicates.

We predict that non-relationship lenders (that is, lenders with no relationship with a given

borrower before CDS initiation) are more likely to serve as lead arrangers or join loan syndicates

as participants once CDSs on a borrower’s debt become available. Non-relationship lenders

competing for a loan face adverse selection risk, as relationship lenders’ prior experience with the

borrower gives them a substantial information advantage (Boot 2000). Because relationship

lenders exploit this advantage and bid for good loans while avoiding bad ones, non-relationship

lenders will compete less aggressively for a loan, which typically results in them losing it to a

better-informed incumbent lender (Rajan 1992). But once CDSs become available, an opportunity

to hedge against potential credit losses should significantly mitigate non-relationship lenders’

adverse selection concerns. CDS spreads also often reveal private information about a borrower

through informed trading by relationship lenders, further diminishing non-relationship lenders’

information disadvantage (e.g., Acharya and Johnson 2007; Qiu and Yu 2012; Batta et al. 2016).

We expect lower adverse selection to increase the willingness of non-relationship lenders to bid

for a loan, thus increasing their probability of winning a deal. This reasoning applies to non-

relationship lenders competing with incumbent lead arrangers to serve as the lead arranger of a

2

loan as well as to non-relationship lenders competing to join the syndicate with other participants

that have superior information about the borrower through prior lending.

Using staggered initiation of CDS trading across CDS firms, we use a difference-in-

differences research design with firm and time fixed effects. We find a positive and significant

effect of CDSs on non-relationship lending for both lead arrangers and syndicate participants. The

probability that a non-relationship lead arranger (syndicate participant) arranges (joins) the loan

syndicate is 1.39 (1.28) times higher following CDS trading initiation on a borrower’s debt. These

findings are robust when we perform short-window analyses around CDS initiation, propensity-

score match CDS firms with non-CDS firms, and use the instrumental variable approach.

To better understand the underlying mechanisms of non-relationship lending decisions

following CDS initiation, we next explore whether non-relationship lenders are more likely to

focus on commercial aspects of the lending relationship. Lending decisions are typically driven by

the overall profitability of the relationship with the borrower, including non-interest income

(Standard & Poor’s 2011). Lenders cross-sell various services, such as investment banking,

derivatives, and structured finance, generally aiming to serve their borrowers as a “one-stop shop”

for all financial services (Drucker and Puri 2005; Yasuda 2005; Fang et al. 2013). Because non-

relationship lenders initiate a lending relationship only after CDSs become available and are

therefore likely to hedge their exposure, thus reducing concerns about borrower credit risk, we

predict that they will pursue new borrowers with high cross-selling potential. Cross-sold products

typically generate substantial fee revenues, enhancing the profitability of the new relationship, but

impose only insignificant pressure on regulatory capital ratios. These fees will also help

compensate for the cost of purchasing CDS protection, further reinforcing our predictions.1

1 Substantial loan origination fees charged by lead arrangers further help them compensate for CDS hedging costs. Origination fees typically range from one to five percent of the total loan commitment, depending primarily on the

3

Because lenders’ expectations about borrowers’ future business needs are difficult to assess,

we use multiple measures of cross-selling potential: (a) media coverage of a borrower’s activities,

including public offerings, mergers and acquisitions (M&A), and investments; (b) a borrower’s

M&A after a loan’s issuance; (c) whether a borrower is in the aviation, shipbuilding, oil, refinery,

gas, or telecommunication industries, which offer lenders opportunities to sell structured finance

products; (d) a borrower’s use of derivatives; and (e) the extent of a borrower’s foreign operations,

which captures transaction banking opportunities and related foreign-currency hedging. For all

measures, we find that, following CDS initiation, lead arrangers are more likely to enter into new

lending relationships if a borrower has high cross-selling potential. The likelihood of non-

relationship participants joining the syndicate is also enhanced by cross-selling potential, but to a

lesser degree, which is likely explained by lead arrangers being the primary beneficiaries of those

opportunities (e.g., Ivashina and Kovner 2011).

We also expect the commercialization of the new lending relationships to be reflected in

lenders’ weaker monitoring incentives. Non-relationship lenders that establish a relationship with

the borrower following the onset of CDS trading should be more likely to rely on CDSs, rather

than on borrower monitoring, to protect themselves against credit risk. Less intensive monitoring

will allow these lenders to reduce monitoring costs, further enhancing the profitability of new

relationships. Weak monitoring may also help a non-relationship lender attract new borrowers, as

well as give these borrowers flexibility to pursue new business activities, which in turn will create

additional cross-selling opportunities. Therefore, although prior studies show that CDSs diminish

the strength of lender monitoring (e.g., Subrahmanyam et al. 2014; Martin and Roychowdhury

2015), we predict this adverse effect to be stronger for non-relationship lenders.

complexity of the transaction (Standard & Poor’s 2011). In 2017, loan origination fees amounted to $11.5 billion in the US syndicated loan market.

4

We base lender monitoring analyses on lead arrangers because they perform the primary

monitoring of the borrowers on behalf of syndicate participants (e.g., Lee and Mullineaux 2004;

Sufi 2007; Ivashina 2009). We measure lead arrangers’ monitoring incentives by the strength of

their control rights, since lenders retain stronger rights when they intend to monitor a borrower

more intensively (Roberts and Sufi 2009a, 2009b; Roberts 2015). Loan contracts with a higher

ratio of performance (income-statement-based) covenants to total financial covenants and those

that incorporate interest-increasing performance-pricing provisions endow lenders with stronger

control rights (Roberts and Sufi 2009a; Christensen and Nikolaev 2012; Christensen et al. 2016).

Consistent with our prediction, we find that, following CDS initiation, non-relationship lead

arrangers impose a lower ratio of performance covenants and are less likely to impose interest-

increasing provisions relative to relationship lead arrangers. Further, we show that, in contrast to

non-relationship lead arrangers, relationship lead arrangers do not relinquish their control rights

after CDS initiation. This evidence provides new insight into the previously documented negative

effect of CDSs on lenders’ monitoring incentives (e.g., Chakraborty et al. 2015; Shan et al. 2015).

We show that this effect stems primarily from lenders that initiate a relationship with the borrower

after CDS initiation, while relationship lenders do not reduce their monitoring efforts.

Lead arrangers also have stronger monitoring incentives when they have more skin in the

game; that is, when they retain a larger loan share (e.g., Sufi 2007; Ivashina 2009). Therefore, we

predict that, following CDS initiation, non-relationship lead arrangers retain a smaller loan share

relative to relationship lead arrangers. We find strong support for this prediction. Importantly,

consistent with covenant and performance-provision tests, we show the opposing responses to

CDS initiation by relationship and non-relationship lenders. Although prior research shows that

lead arrangers retain larger loan share in the post-CDS period (e.g., Amiram et al. 2017), we find

5

that this effect is driven primarily by relationship lenders, while non-relationship lenders do not

increase their skin in the game.

We complement these analyses by exploring whether the characteristics of non-relationship

lenders differ in the post- versus the pre-CDS initiation period. If CDS-protected non-relationship

lead arrangers are less committed to monitoring, we predict that there is a higher likelihood that

lead arrangers with lower monitoring efficiency will initiate relationships in the post period.

Because greater distance to the borrower impedes a lender’s information gathering and increases

monitoring costs, we measure monitoring efficiency by the distance between the lead arranger and

the borrower and by whether the lead arranger is a foreign institution (e.g., Berger et al. 2005;

Degryse and Ongena, 2005; Agarwal and Hauswald 2010). Following prior research, we also view

reputable lead arrangers as having stronger monitoring ability (e.g., Sufi 2007; Bushman and

Wittenberg-Moerman 2012). We find that, relative to relationship lenders, non-relationship lenders

are likely to be more distant from borrowers, foreign, and less reputable once CDSs become

available, consistent with such lenders being less likely to engage in costly monitoring.

We acknowledge that the stronger negative influence of CDSs on the monitoring of non-

relationship lead arrangers should negatively affect syndicate participants’ willingness to join the

syndicates they arrange. Prior studies suggest that agency problems within a syndicate are

mitigated if participants with prior relationships with the borrower join the syndicate, since they

are more informed about the borrower’s performance and creditworthiness (e.g., Sufi 2007;

Ivashina 2008). Thus, although we hypothesize and find that non-relationship participants are more

likely to join the syndicate following CDS initiation, we predict that this effect will be attenuated

if a non-relationship lead arranger syndicates a loan. We indeed find that, following CDS initiation,

the probability that a non-relationship participant joins a syndicate is lower if the loan is arranged

6

by a non-relationship relative to a relationship lead arranger.

We conduct additional exploratory tests of the effect of CDSs on non-relationship lending.

To disrupt existing lending relationships, non-relationship lenders may entice borrowers with

favorable loan terms, such as a low interest rate and large loan size. However, because hedging via

CDSs is costly, more favorable terms may diminish the benefits of the new lending relationship.

We find that, following CDS initiation, non-relationship lead arrangers do not charge lower

spreads but do issue smaller loans, relative to relationship lead arrangers. Coupled with the

significant impact of cross-selling potential on non-relationship lenders’ decisions, this evidence

suggests that, even though these lenders provide new borrowers with less credit in order to reduce

hedging costs, they still get a mandate to provide the borrowers with additional financial services.

To shed additional light on lenders’ incentives to initiate new lending relationships following

CDS initiation, we explore non-relationship lenders’ performance. Lenders often experience low

loan growth due to internal credit risk-management restrictions or regulatory capital constraints.

Since CDSs offer more flexible credit risk management and alleviate capital constraints, we predict

that lenders with low loan growth are more likely to seek new borrowers after the onset of CDS

trading. We also expect lenders with low profitability to have stronger incentives to initiate new

relationships once they can hedge their exposure through CDSs, as these relationships may boost

profitability by generating additional interest revenue and cross-selling fees. Consistent with these

predictions, we find that low loan growth and profitability amplify the effect of CDSs on the

probability of initiating new relationships for both lead arrangers and syndicate participants.

Our paper contributes to the literature on the influence of the CDS market on private lending.

Prior studies find a significant impact of CDSs both on borrowers’ access to credit and on the

strength of lenders’ monitoring (e.g., Ashcraft and Santos 2009; Saretto and Tookes 2013;

7

Subrahmanyam et al. 2014; Martin and Roychowdhury 2015). These studies, however, mostly do

not differentiate between existing and new lending relationships.2 We extend this literature by

exploring how CDSs affect the formation and commercialization of new relationships. Further, we

supplement prior studies that examine the underlying mechanisms through which CDSs affect

lenders’ monitoring incentives, such as lead arrangers’ loan shares and control rights (e.g.,

Chakraborty et al. 2015; Shan et al. 2015; Amiram et al. 2017). We show that the influence of CDS

trading on these mechanisms depends to a large extent on whether lenders had an established

relationship with a borrower before CDS initiation.

We also contribute to the large literature on relationship lending. While some studies show

important benefits to borrowers from established lending relationships (e.g., Petersen and Rajan

1994, 1995; Berger and Udell 1995; Brahrath et al. 2009), others emphasize its dark side, where

borrowers become locked in a relationship due to the information advantage of incumbent lenders

(e.g., Sharpe 1990; Rajan 1992). We provide evidence consistent with CDS trading fundamentally

altering competition in the loan market by enhancing the willingness of non-relationship lenders

to compete for new borrowers. Relatedly, our findings underscore important implications of

financial innovation for the evolution of the lender-borrower relationship.

2 In a concurrent paper, Shan et al. (2016) also document the higher likelihood of loans syndicated by non-relationship lead arrangers following CDS initiation. They attribute this finding to borrowers switching to new lead arrangers because their current borrower–lead-arranger relationship is compromised by CDSs and becomes less valuable to the borrower. In contrast, building on Rajan (1992), we suggest that CDSs decrease adverse selection between incumbent and non-relationship lenders, thus increasing the latter’s willingness to compete for new borrowers and consequently the likelihood of winning a loan deal. We also focus on the commercialization of the new lending relationship in the post-CDS period, as reflected by non-relationship lenders pursuing borrowers with higher cross-selling potential and by their weaker monitoring incentives, which Shan et al. (2016) do not explore. Note, too, that our findings with respect to non-relationship lead arrangers’ monitoring incentives further undermine Shan et al.’s (2016) motivation. They argue that borrowers switch away from relationship lenders primarily because they are concerned about these lenders’ weaker monitoring incentives. In contrast, we show that relationship lenders actually do not relinquish their control rights and do increase their “skin in the game” following CDS initiation, while non-relationship lenders monitor substantially less intensively than relationship lenders.

8

Finally, we add to the growing literature on the importance of noncredit revenue sources in

private lending. Prior studies show that lenders gain future underwriting of a borrower’s bonds and

equity (e.g., Drucker and Puri 2005; Yasuda 2005). There is also evidence that lenders price into

the loan spread a borrower’s cross-selling potential as well as its board network, as this network

offers an advantage in cross-selling services to other firms (e.g., Ivashina and Kovner 2011; Zhao

2017). We complement these studies by showing that a borrower’s cross-selling potential is

instrumental in lenders’ decisions to initiate new relationships. We also show that CDS protection

leads to the substantial commercialization of these relationships.

The next section presents hypothesis development. Section 3 describes data and sample

selection. Section 4 reports our findings and Section 5 concludes.

2. Related Literature and Hypothesis Development

2.1 CDS trading and non-relationship lending in the syndicated loan market

The introduction of CDSs significantly influenced the private debt market by offering banks

hedging opportunities to lay off borrowers’ credit risk, while maintaining lending relationships

(e.g., Saretto and Tookes 2013). Even if lenders do not hedge their loan exposure at loan initiation,

the existence of CDSs offers them a liquid resale option and thus increases their willingness to

extend credit. Further, CDSs allow lenders more flexible risk management and provide them an

opportunity to reduce regulatory capital requirements by substituting the risk weight of the CDS

counterparty (typically, a large financial institution) for that of the borrower (Saretto and Tookes

2013; Streitz 2015; Martin and Roychowdhury 2015; Shan et al. 2016).

Yet CDSs also induce empty creditor problems as CDS-protected lenders become

intransigent in debt renegotiation or even push borrowers into inefficient liquidation (Hu and Black

2008; Bolton and Oehmke 2011). CDSs reduce lenders’ monitoring incentives because hedging

9

their loan exposure may be less costly alternative to protect themselves against default risk than

intensive information collection and monitoring. Consistent with weaker monitoring, prior studies

find that, following CDS initiation, the cost of debt increases for risky and informationally opaque

borrowers, borrowers experience more credit rating downgrades and bankruptcies and report less

conservatively, and lenders impose looser financial covenants and weaker restrictions after

covenant violations (Ashcraft and Santos 2009; Subrahmanyam et al. 2014; Martin and

Roychowdhury 2015; Chakraborty et al. 2015; Shan et al. 2015). Amiram et al. (2017) show that

syndicate participants, who largely delegate monitoring to the lead arranger, require the arranger

to have more skin in the game to compensate for the reduction in its monitoring incentives after

the initiation of CDS trading.

We extend these studies by examining how the availability of CDSs affects the likelihood of

lenders without a prior relationship with the borrower serving as the lead arranger or joining its

loan syndicate as a participant. Because relationship lenders have an information advantage, due

to their extensive knowledge of a borrower’s operations and creditworthiness, non-relationship

lenders competing for a borrowers’ loans face substantial adverse selection risk (Boot 2000). This

typically results in a non-relationship lender losing competition for a borrower’s loan to a better-

informed incumbent lender. Specifically, Rajan (1992) shows that relationship lenders exploit their

information advantage and bid for good loans while avoiding bad ones. Because relationship

lenders avoid bad loans, non-relationship lenders face significant losses if they do bid. As a result,

non-relationship lenders, being aware of the adverse selection problem, compete less aggressively,

which reduces the probability that they win the loan.

The availability of CDSs on a borrower’s debt mitigates adverse selection risk that non-

relationship lenders face because it allows them to hedge loan exposure, thus assuaging their

10

concerns regarding potential credit losses when initiating a new relationship. CDS spreads also

often reveal private information about a borrower, ahead of public disclosures and price discovery

in other markets, largely due to insider trading of relationship lenders (e.g., The Financial Times

2005; Acharya and Johnson 2007; Qiu and Yu, 2013; Batta 2016). The revelation of a substantial

private information through CDSs should mitigate non-relationship lenders’ information

disadvantage, further alleviating their adverse selection concerns.

This reasoning applies to lead arrangers as well as syndicate participants. Relationship

lenders and non-relationship lenders compete to serve as the lead arranger on a borrower’s loan.

Lower adverse selection, due to CDS availability, is expected to increase non-relationship lead

arrangers’ willingness to compete for a borrower’s loan, increasing the probability of their winning

the deal (Rajan 1992). Similarly, lenders compete to join syndicates. Syndication represents an

auction in which loan participants submit sealed bids to the lead arranger, and the number of

invitations extended to potential participants typically exceeds the number of lenders who will

fund the loan (Champagne and Kryzanowski 2007; Ivashina and Sun 2011). Lenders with superior

information about the borrower, because of prior lending relationships, are likely to pursue

participation in good loans, leaving bad ones to non-relationship participants. We expect lower

adverse selection, due to CDS hedging and the revelation of private information through CDS

spreads, to increase the willingness of non-relationship lenders to bid for loan participation.

Building on these argument, we state our first hypothesis as follows:

H1: The probability that a non-relationship lead arranger (syndicate participant) arranges

(joins) the loan syndicate is higher following the initiation of CDS trading on a borrower’s debt.

2.2 The Commercialization of New Lending Relationships

We next investigate whether a borrower’s cross-selling potential intensifies the effect of CDS

11

trading on non-relationship lending in the syndicated loan market. Lenders invest in loans for more

than just the interest income; they consider the overall profitability of the relationship, including

noncredit revenue (Standard & Poor’s 2011). In particular, lenders typically adopt a portfolio-

management approach, in which they allocate capital to borrowers based on the total return

generated by a relationship, relative to its risk.

Lenders cross-sell various services, including investment banking, derivatives, and

structured finance. Drucker and Puri (2005) and Yasuda (2005) show that serving as a lead arranger

helps a bank gain future underwriting of a borrower’s bonds and equity. Ivashina and Kovner

(2011) also show that lenders account for cross-selling potential when pricing leveraged buyout

loans. Fang et al. (2013) provide evidence consistent with banks aiming to serve their borrowers

as “one-stop shop” for financial services.

Because non-relationship lenders initiate lending relationship only once CDSs become

available, they are likely to hedge their exposure, thus diminishing concerns about a borrower’s

credit risk. Thus, we expect CDS-protected non-relationship lenders to focus more on commercial

aspects of the lending relationship. We predict that these lenders will pursue new borrowers with

high cross-selling potential. Cross-sold products are mostly fee-based and entail minimal risk-

weighted assets. Therefore, cross-selling will generate substantial fee revenues for the lender,

while imposing only insignificant pressure on regulatory capital ratios. Fees generated by cross-

selling will also help compensate for the cost of purchasing CDS protection. Note that, although

non-relationship lenders would have benefited from a lending relationship with high cross-selling

potential borrower even before CDS initiation on the borrower’s debt, they were reluctant to

initiate such relationships, likely due to the high adverse selection they faced without credit

protection via CDS. We therefore hypothesize:

12

H2: The increase in the probability that a non-relationship lead arranger (syndicate

participant) arranges (joins) the loan syndicate following the initiation of CDS trading on a

borrower’s debt is higher when the borrower has high cross-selling potential.

We also expect that the commercialization of the lending relationships initiated following

the onset of CDS trading leads to the reduction in lender monitoring incentives. Prior studies

already document that CDSs undercut lenders’ monitoring incentives (e.g., Ashcraft and Santos

2009; Subrahmanyam et al. 2014; Martin and Roychowdhury 2015; Amiram et al. 2017), but we

predict that this effect is stronger for non-relationship lenders. Because non-relationship arrangers

initiate new relationships with the borrowers only once CDSs become available, they are more

likely to purchase CDS protection than to engage in intense borrower monitoring to address credit

risk concerns. Lenders will bear lower monitoring costs with weaker borrower monitoring, thus

boosting the profitability of new relationships. Moreover, monitoring less intensively may help

non-relationship lead arrangers attract new borrowers, as borrowers prefer fewer constraints on

their actions (e.g., Berlin and Mester 1992; Bradley and Roberts 2015). Allowing borrowers

greater flexibility to engage in new business activities should also create additional cross-selling

opportunities, further benefiting non-relationship lenders. Because it is lead arrangers that assume

the primary monitoring responsibilities on behalf of syndicate participants and negotiate loan terms

with the borrower (e.g., Lee and Mullineaux 2004; Sufi 2007; Ivashina 2009), we focus our third

hypothesis on the lead arrangers of syndicates:

H3: Following CDS initiation, non-relationship lead arrangers have weaker monitoring

incentives, relative to relationship lead arrangers.

13

3. Sample, Data, and Descriptive Statistics

3.1 Data sources and sample selection

We obtain syndicated loan characteristics from DealScan and CDS data from Markit.

Borrowers’ characteristics are from Compustat and CRSP, and banks’ characteristics are from Call

Reports (Form Y-9C regulatory filings). We collect firm credit ratings from Compustat and

Mergent FISD, data on analyst coverage from I/B/E/S, data on media articles from RavenPack

News Analytics, and data on mergers and acquisitions from SDC.

We start sample selection with all syndicated loans issued from 1994 to 2015, as reported in

DealScan. There are 46,061 loan packages (deals) issued to US firms over this period. We match

this sample to Compustat and eliminate observations with insufficient loan and firm data, resulting

in 29,559 loan packages, issued to 6,956 firms. We match this sample to Markit and identify firms

with traded CDSs at any point over our sample period (CDS firms hereafter). Following prior

literature (e.g., Ashcraft and Santos 2009; Amiram et al. 2017), we identify the date of CDS trading

initiation as the first date when a CDS quote for a firm is provided by Markit.3 In line with Amiram

et al. (2017), we eliminate firms for which the first CDS trade date falls in January 2001, the first

month of Markit’s coverage, because of the ambiguity of the exact date of initiation for these firms.

We also require each CDS firm to have at least one loan issued in both the pre- and post-CDS

initiation periods and non-CDS firms to have at least one loan issued starting in 2001, when CDS

firms in our sample begin to have traded CDSs. Our final sample contains 843 CDS firms and

2,814 non-CDS firms, corresponding to 23,316 loan packages. This process yields 27,547

(177,936) arranger-loan (participant-loan) observations.

3 Our main findings are unchanged when we identify CDS initiation date as the first date when the quote for a five-year USD-denominated CDS contract is provided by Markit, following Ashcraft and Santos (2009). These contracts are the most commonly traded in the CDS market.

14

3.2 Descriptive statistics

Table 1 presents descriptive statistics of the main variables used in our test. The mean value

of No Relationship - Lead Arranger indicates that 38% of lead arrangers in our sample do not have

a prior relationship with the borrower. For lead arrangers that syndicate loans to CDS firms

following CDS initiation, we define the No Relationship - Lead Arranger indicator variable to be

equal to 1 if the lead arranger has not syndicated a borrower’s loans before the CDS initiation date

and 0 otherwise. For lead arrangers that syndicate loans to CDS firms prior to CDS initiation or to

non-CDS firms, we define No Relationship - Lead Arranger to be equal to 1 if the lead arranger

has not syndicated a borrower’s loans before the issuance date of the loan under consideration and

0 otherwise. The mean value of No Relationship - Participant indicates that 44% of syndicate

participants in our sample do not have prior relationships with the borrower. For syndicate

participants in loans to CDS firms following CDS initiation, we define the No Relationship -

Participant indicator variable to be equal to 1 if the participant has not participated in a borrower’s

loans before the CDS initiation date and 0 otherwise. For participants in loans to CDS firms prior

to CDS initiation or to non-CDS firms, we define No Relationship - Participant to be equal to 1 if

the participant has not participated in a borrower’s loans before the issuance date of the loan under

consideration and 0 otherwise.4

For both the lead arranger and participant samples, the firms are relatively large, as reflected

by the mean values of Assets, defined as the natural logarithm of total assets (detailed variable

definitions are reported in Appendix A). The average ratio of net income to total assets (ROA) is

4 We acknowledge that defining No Relationship based on the CDS initiation date for CDS firm and based on loan issuance date for non-CDS firms may raise a concern about the validity of our findings. In Section 4.2, we use difference-in-differences approach, where we assign a pseudo-CDS initiation date to the matched non-CDS firms using the initiation date of the paired CDS firms. This allows us to measure No Relationship over the same period for each pair of CDS and matched non-CDS firms.

15

0.03 for both samples, and 21% (16%) of sample observations relate to firms experiencing losses

(Loss) for the lead arranger (participant) samples. The mean leverage ratio (Leverage), measured

by the ratio of total liabilities to total assets, is 0.64 (0.66), and the average interest coverage ratio

(Interest coverage), measured as EBIT divided by the interest expense, is 11.02 (9.65), while mean

asset tangibility (Tangibility), measured as property, plant, and equipment divided by total assets,

is 0.60 (0.59) for the lead arranger (participant) sample. In addition, 58% (72%) of the lead arranger

(participant) sample observations relate to borrowers that are rated (Rated) and 33% (42%) to

borrowers with investment-grade ratings (Investment Grade). We also report statistics for loan

characteristics that serve as controls. For both samples, the mean and median value of loan size

(Amount), measured by natural logarithm of the loan amount, suggest that sample observations

relate to loans that are relatively large (Amount), have an average maturity (Maturity) of

approximately four years, and are subject to less than two financial covenants (#Covenants). A

majority of observations are characterized by performance pricing provisions (PP), and 10%

(11%) of them have a loan guarantor (Guarantor) for the lead arranger (participant) sample.

4. Empirical Results

4.1 The effect of CDS Trading on Non-Relationship Lending

To examine our first hypothesis that lenders are more likely to issue loans to a new borrower

following the CDS trading initiation on the borrower’s debt, we estimate the following logit model.

No Relationship = β 0 + β1POST + β2Assets + β3ROA + β4Loss + β5Leverage

+ β6Interest Coverage + β7Tangibility + β8Rated

+ β9Investment Grade + β10Amount + β11Maturity + β12Guarantor

+ γFixed Effects + ε,

(1)

16

where No Relationship is one of the two indicator variables — No Relationship-Lead Arranger

and No Relationship-Participant — defined as previously. Our main variable of interest is the

POST indicator variable, which equals to 1 if the loan is issued after the CDS trading initiation

date and 0 otherwise (this variable takes the value of 0 for all loans to non-CDS firms). If non-

relationship lending increases after CDS trading initiation on a borrower’s debt, we expect a

positive and significant coefficient on POST.

We control for firm and loan characteristics that can affect lenders’ decisions to initiate new

lending relationships, including a firm’s size, profitability, the incidence of losses, interest

coverage, tangibility and credit rating characteristics, as well as loan size, maturity and whether a

loan has a guarantor. All firm characteristics are defined as previously and measured in the year

preceding a loan’s issuance. For loan packages that contain more than one loan (facility), we follow

Ball et al. (2008) and Ivashina (2009) and use characteristics of the largest one. Our findings are

unchanged when we control for the weighted-average loan characteristics of all loans in the

package, where weights are based on loan (facility) size (untabulated).5

To aid identification strategy, we include in Model (1) firm and year fixed effects, allowing

us to implement a difference-in-differences research design, because CDS initiation dates are

staggered across CDS firms. Firm fixed effects control for the time-invariant differences between

CDS and non-CDS firms, while year fixed effects control for the time-varying factors common to

all sample firms (e.g., Bertrand and Mullainathan 2003; Sapienza 2002; Valta 2012; Amiram et al.

2017).6 We acknowledge that, because firms may issue loans in months both before and after the

5 We do not control for the number of covenants and performance pricing provisions, as these characteristics are typically determined during the loan negotiation and therefore cannot affect lenders’ choice of whether to start a new lending relationship. In any case, in untabulated analyses, we find that our results are robust to the inclusion of these variables. 6 In section 4.2 below, we propensity-score match CDS firm to non-CDS firm, which allows us to assign the CDS initiation date to each matched non-CDS firm. Our results are robust to this standard difference-in-difference approach.

17

initiation date during the CDS initiation year, employing year-month fixed effects is a better

strategy to identify initiation effects. Due to concerns regarding a large number of fixed effects in

nonlinear models (e.g., Maddalla 1987; Greene 2004), as an additional specification, we estimate

Model 1 with a linear probability model, where we substitute year fixed effects with year-month

fixed effects. Moreover, as many of our subsequent analyses incorporate interaction terms, linear

probability estimations also mitigate concerns regarding the interpretation of interaction terms in

our nonlinear model estimations (e.g., Norton et al., 2004). To correct for within-firm correlation

in the error term, we cluster standard errors at the firm level.

We present our findings in Panel A of Table 2. In columns 1 and 2, we report the estimation

of Model (1) employing Logit and OLS models, respectively, for the lead arranger sample. We

find a positive and significant coefficient on POST for both Logit and OLS specifications.

Economically, based on the OLS specification, the probability that a non-relationship lead arranger

syndicates the loan is 1.39 times higher following the CDS trading initiation (note that we measure

economic significance based on the OLS specification for all tests). We find similar results for

syndicate participants, as reported in columns 3 and 4. The probability that a participant with no

relationship with the borrower joins the syndicate is 1.28 times higher once CDSs become

available. These results are consistent with our primary prediction that lenders are more likely to

initiate a new lending relationship after the inception of CDSs referencing the borrower’s debt.7

With respect to controls, the negative and significant coefficient on Leverage suggests that

higher leverage deters non-relationship lenders. The positive and significant coefficient on

Tangible for the lead arranger sample implies that higher asset tangibility attracts non-relationship

arrangers. When borrowers issue longer maturity loans and those with guarantors, we find that it

7 In untabulated analyses, we find that the effect of CDS trading on non-relationship lending is more pronounced for lead arrangers than syndicate participants.

18

is more likely for these loans to be arranged by non-relationship lenders.

One concern that might arise when using our difference-in-differences research design is

whether the parallel trend assumption holds. To address it, we augment Model (1) with an indicator

variable Last Loan, which is equal to 1 for the last loan issued before the CDS initiation date and

0 otherwise. If there exists a pre-tend of non-relationship lending before CDS initiation, we expect

to find a higher likelihood of non-relationship lending for the last loan before initiation (Heider

and Ljungqvist 2015; Amiram et al. 2017). As we report in Panel B, the coefficients on Last Loan

are insignificant and negative across all model specifications, indicating that the likelihood of non-

relationship lending is not higher for the last loan before CDS initiation date relative to the loans

preceding this loan. This evidence suggests that the parallel trend assumption holds in our tests.

To further verify the robustness of our findings, we repeat our analyses using a short-window

sample that limits observations of CDS firms to three years before and after CDS initiation. These

analyses mitigate concerns that factors other than CDS initiation could drive our results. We report

these tests in Panel C. Consistent with our main findings, the coefficients on POST is positive and

significant for all lead arranger and participant specifications.

We provide additional support for our inferences by examining how the effect of CDSs on

non-relationship lending varies with a borrower’s information opacity. We expect hedging through

CDSs and private information revealed by CDS spreads to be more valuable for non-relationship

lenders when they lend to opaque firms, as adverse selection costs significantly increase with a

borrower’s opacity. To measure information opacity, we rely on a borrower’s analyst coverage,

which is helpful for lenders in assessing borrower creditworthiness (Güntay and Hackbarth 2010;

Mansi et al., 2011). Low Coverage is an indicator variable equal to 1 if the number of equity analyst

following the borrower is below the sample median in a year of a loan’s issuance and 0 otherwise.

19

We augment model (1) with Low Coverage and the interaction term POST × Low Coverage. As

we report in Appendix B, Table B1, the coefficient on POST × Low Coverage is positive and

significant for the lead arranger specifications, suggesting that the effect of CDSs on the likelihood

of a non-relationship lead arranger syndicating a loan is more pronounced for opaque borrowers.

Following CDS initiation, lead arrangers are 1.21 times more likely to initiate a new lending

relationship with opaque borrowers relative to more transparent borrowers. 8 However, the

coefficient on POST × Low Coverage is positive but insignificant for the participant specifications.

This evidence suggests that, because lead arrangers perform the primary screening and monitoring

of the borrower, they are likely to incur higher adverse selection costs when initiating a new

relationship with an opaque firm.

4.2 Addressing Endogeneity Concerns

Although Model (1) includes firm and time fixed effects, we next employ PSM and IV

approaches to further assuage endogeniety concerns associated with fundamental differences

between CDS and non-CDS firms. To construct a matched sample of CDS and non-CDS firms for

PSM analysis, we employ the CDS initiation model following Martin and Roychowdhury (2015).

CDS Initiation = β0 + β1Rated + β2Investment Grade + β3Assets + β4Leverage

+ β5Profit Margin + β6Return Volatility + β7MTB + ε,

(2)

where CDS Initiation is an indicator variable equal to 1 for the year of CDS initiation and 0

8 Although our findings with respect to the coefficient on the POST × Low Coverage interaction term are consistent across Logit and OLS specifications, to further address the concern regarding the interpretation of interaction terms in non-linear models, we apply Buis’ (2010) methodology. We continue to find positive and significant coefficient on POST × Low Coverage (untabulated). We perform this robustness test for all other analyses that include interaction terms in non-linear models. We find that the direction and economic significance of coefficients on interaction terms remains the same as in non-linear model estimations that we tabulate.

20

otherwise. Rated, Investment Grade, Assets, and Leverage are as previously defined. Profit Margin

is net income divided by sales. Return Volatility is the standard deviation of the firms’ monthly

stock return over the year. MTB is the market value divided by the book value of equity. All

variables are measured in the year before the year under consideration. We tabulate the estimation

results of the first stage PSM model in Appendix B, Table B2, Panel A. In line with prior studies

(Martin and Roychowdhury 2015; Kim et al. 2018), we find that the probability of CDS initiation

is higher when a firm is rated, has an investment grade rating, and is larger. This probability is

lower when a firm is more profitable.

Using the propensity score (“predicted probability of CDS initiation”) estimated by the first-

stage PSM model (2), we construct the one-to-one matched PSM sample by selecting for each

CDS firm a non-CDS firm having the closest propensity score. The matched sample enables us to

employ a standard difference-in-difference research design, because we can assign a pseudo-CDS

initiation date to the matched non-CDS firms using the initiation date of the paired CDS firms. We

allow non-CDS firms to be matched to multiple CDS firms to improve covariate balancing and

retain sample size (e.g., Saretto and Tookes 2013). Our final sample contains 719 CDS firms and

395 matched non-CDS firms (referring to 161,831 lender-loan observations). We tabulate the

results of the covariate balancing analyses in Panel B of Table B2. These analyses suggest that the

PSM matched samples are well balanced across the covariates for most CDS initiation

determinants. The CDS firms are larger than matched non-CDS firms at the 10% significance

level, but, economically, the difference in firm size is relatively small, given the absolute firm size

of the matched CDS and non-CDS firms (mean assets for CDS (non-CDS) firms are USD 4,447

million (USD 3,944 million).

Panel A of Table 3 presents the estimation results based on the PSM sample. We augment

21

Model (1) with POST × CDS Traded interaction term, where CDS Traded is equal to 1 for CDS

firms and 0 for non-CDS firms (the CDS Traded indicator variable is subsumed by firm fixed

effects). The coefficient on POST × CDS Traded is a difference-in-differences estimator and is

expected to be positive. Consistently, we find positive and significant coefficients on POST × CDS

Traded for both lead arranger and syndicate participant specifications, supporting our main

hypothesis that non-relationship lenders are more likely to extend loans to new borrowers once

CDSs become available. Economically, a lead arranger (participant) is 1.17 (1.13) times more

likely to initiate non-relationship lending following CDS initiation. To further examine the

robustness of our findings, we conduct PSM analysis for a short-window sample that limits

observations to three years before and after CDS initiation date (pseudo-initiation date for non-

CDS firms). The results presented in Panel B corroborate our prior findings: coefficients on POST

× CDS Traded are significantly positive in all specifications.

In addition to addressing endogeneity, the analyses in Table 3 mitigate concerns regarding

the measurement of non-relationship lending for CDS and non-CDS firms in our primary tests.

While for loans of CDS firm issued following CDS initiation we measure lending relationship

relative to the CDS initiation date, because this date cannot be defined for non-CDS firms, we

define non-relationship lending for these firms based on whether a lender had a relationship with

the firm before the current loan’s issuance date. By assigning a pseudo-CDS initiation date to non-

CDS firms, a standard difference-in-differences approach allows us to measure non-relationship

lending over the same period for each pair of CDS and matched non-CDS firms.

Next, we perform additional analysis based on the IV approach to further address

endogeneity concerns. Following prior literature, we use lenders’ foreign exchange (FX) derivative

position as an instrument for CDS trading (e.g., Saretto and Tookes 2013; Subrahmanyam et al.

22

2014; Shan et al. 2015; Amiram et al. 2017). Because banks that hedge one component of their

portfolio are more likely to hedge others (e.g., Minton et al. 2009), we expect banks that hedge

foreign exchange risk to also hedge credit risk via CDSs. Further, lenders’ FX derivatives position

is a macro, rather than a firm-specific, hedge (e.g., Saretto and Tookes 2013). Thus, while related

to a bank’s general propensity to hedge, the decision to hedge foreign exchange should be

exogenous to a decision to start a lending relationship with a particular borrower.

To construct the Forex instrument, we collect notional amounts of FX derivatives from Call

Reports filed by bank holding companies. For banks with available data, Forex is defined as the

ratio of notional amounts of FX derivatives to total assets. We augment CDS initiation model (2)

with Forex and present, in Panel C, the results of the second-stage IV regression estimated

simultaneously with the first-stage CDS initiation model. For both specifications, the coefficients

on Post are positive and significant. Although we cannot be certain that the Forex instrument is

truly exogenous, the results of IV tests confirm that lead arrangers and participants are more likely

to enter into a new lending relationship when CDSs are available on the borrower’s debt.9

4.3 Borrower Cross-Selling Potential

In this section, we examine whether non-relationship lenders are more likely to pursue

borrowers with higher cross-selling potential once CDSs become available. These analyses will

help illuminate whether CDS-protected lenders pursue commercial aspects of the lending

relationship when they initiate a new relationship. Since lenders’ expectations about borrowers’

future business needs are not directly observable, we employ multiple measures of high cross-

9 Due to Stata computational limitations, we cannot estimate the second-stage IV regression simultaneously with the first-stage model (2) when the second stage is estimated as the non-liner model with firm and year fixed effects or an OLS model with firm and month-year fixed effects. Therefore, to perform IV tests, we estimate the second-stage model as OLS with firm and year fixed effects.

23

selling potential to test our prediction.

Our first measure of cross-selling potential is based on the media coverage of borrowers’

business activities, as media is an important information source to non-relationship lenders

(Bushman et al. 2016). We consider media articles that cover topics such as “public offering,”

“merger,” “acquisition,” and “investment” to relate to cross-selling opportunities. We focus on

these topics because lenders benefit from cross-selling security underwriting and M&A advisory

services, and receive significant originating fees if they syndicate loans to finance future

investments of the borrower (e.g., Drucker and Puri 2005; Yasuda 2005; Ivashina and Kovner

2011; Fang et al. 2013; Euromoney 2014). Our discussions with loan officers at several banks

suggest that lenders continuously monitor information in the media to gather information about

borrowers’ projected business activities and recent events, as these events may also indicate future

business opportunities. Therefore, we consider media articles to be related to a borrower’s cross-

selling potential if they cover both past and expected borrower activities.

For each borrower in our sample, we obtain media articles from Ravenpack, which covers

all news disseminated via Dow Jones Newswires. We limit media data to full-size articles with a

relevance score of 75 and above. The relevance score is assigned by RavenPack to indicate when

the firm is strongly related to the underlying news story (the scores range from 0 (low relevance)

to 100 (high relevance)). We identify articles as cross-selling related if their news type, as assigned

by Ravenpack, is public offering, merger, acquisition, or investment. We define the High CS

Potential-Media indicator variable as equal to 1 if media articles about a borrower cover cross-

selling related topics within three years before a loan’s issuance date and 0 otherwise. We then

augment Model (1) with High CS Potential-Media and the interaction term POST × High CS

Potential-Media. Panel A of Table 4 presents our findings (we restrict these analyses to the post-

24

2000 period because of Ravenpack’s data availability). Consistent with our prediction that

following CDS initiation lead arrangers are more likely to enter into a new lending relationships if

a borrower has higher cross-selling potential, we find the positive and significant coefficient on

POST × High CS Potential-Media for the arranger specifications. Economically, non-relationship

arrangers are 1.12 times more likely to syndicate loans to new borrowers with media coverage of

cross-selling opportunities relative to borrowers without such coverage. The coefficient on POST

× High CS Potential-Media is insignificant in participant specifications. This finding is likely to

be explained by lead arrangers being the primary beneficiaries of cross-selling opportunities (e.g.,

Ivashina and Kovner 2011), as they maintain the direct relationship with the borrower.

Because M&A is a major line of business for investment banks, which offers lucrative

business opportunities, such as M&A advisory and related financial services (Liaw 2011), our

second measure of cross-selling potential further addresses a borrower’s M&A activity. Although

our media-based cross-selling potential measure reflects publicly available information about these

deals, lenders’ may also have private information concerning a borrower’s future M&A

transactions. Because we cannot capture lender’s private information, we focus on actual M&A

deals that occurred after loan issuance.

We define the High CS Potential-M&A indicator variable to be equal to 1 if a borrower

engages in M&A deals within three years following a loan’s issuance date and 0 otherwise. We

then augment Model (1) with High CS Potential-M&A and the interaction term POST × High CS

Potential-M&A. We report the results in Panel B of Table 4. Similar to our findings in Panel A,

the coefficient on POST × High CS Potential-M&A is positive and significant for the lead arranger

specifications. Economically, following CDS initiation, non-relationship lead arrangers are 1.23

times more likely to issue loans to new borrowers with M&A business opportunities relative to

25

borrowers without such opportunities. We do not find that cross-selling potential influences

whether non-relationship syndicate participants join the syndicate.

Our third measure of high cross-selling potential is based on a borrower’s industry, as certain

industries offer unique opportunities related to structured finance. In particular, the aviation,

shipbuilding, oil, refinery and gas, and telecom industries offer banks such opportunities, which

have higher fees/margins and lower risk due to high asset collateralization (Oliver Wyman 2006).

Lenders also enjoy additional cross-selling benefits by selling to borrowers in these industries

related derivatives products, such as commodity derivatives (especially for oil, gas, and natural

resources), interest rate derivatives, and foreign-currency derivatives. We define the industry-

based measure of cross-selling potential — High CS Potential-Industry — to be equal to 1 if a

borrower is in one of the listed industries and 0 otherwise.10 As we report in Panel C, the

coefficient on the interaction term POST × High CS Potential-Industry is positive and significant

for the lead arranger specifications, but is insignificant for the participant specifications (note that

High CS Potential-Industry indicator is subsumed by firm fixed effects). Following CDS initiation,

lead arrangers are 1.23 times more likely to start a new lending relationship with borrowers in

industries with high cross-selling opportunities relative to borrowers in other industries. These

findings reinforce our inference that, post CDS initiation, non-relationship lead arrangers are more

likely to initiate lending relationship with borrowers that offer high cross-selling opportunities.

Our fourth measure of cross-selling potential is based on borrowers’ use of derivative-

related products, which also provide lenders with lucrative cross-selling business (e.g., Oliver

Wyman 2006; Euromoney 2014). Derivative transactions offer high margins, as lenders charge

borrowers significant fees while entering into offsetting positions with other counterparties, which

10 We report in Appendix B, Table B3, four-digit SIC codes we used to classify these industries.

26

are typically less expensive than what borrowers are charged. Furthermore, derivative positions

translate into lower risk-weighted assets, as actual exposures are much smaller than the notional

amounts, allowing lenders to reserve less capital for these transactions (e.g., Neilson et al. 2017).

To proxy for borrower future derivative needs, we presume that borrowers with high demand for

derivatives products are likely to have more derivative-related words in their 10-K filings. We

define the High CS Potential-Derivatives variable to be equal to 1 if the ratio of derivative-related

word counts (e.g., “derivative,” “hedging”) to total words count in the 10-K filing in the year

preceding a loan’s issuance is greater (less) than the sample median and zero otherwise. We find

that the coefficient on POST × High CS Potential-Derivatives is positive and significant for both

lead arrangers and participant specifications. Post CDS initiation, lead arrangers (participants) are

1.14 (1.12) times more likely to initiate lending relationships with borrowers having higher

demand for derivative products relative to other borrowers (Panel D).

Our final measure of cross-selling potential addresses the extent of a borrower’s foreign

operations. Foreign operations provide lenders with transaction banking opportunities (e.g., trade

finance) as well as related foreign-currency hedging demand from borrowers. We measure the

extent of foreign operations by an indicator variable High CS Potential-Foreign, which is equal to

1 if the ratio of a number of a borrower’s foreign operating segments to its total number of

operating segments in the year preceding a loan’s issuance is greater than the sample median and

zero otherwise. As we report in Panel E, the coefficient on POST × High CS Potential-Foreign is

positive and significant for all specifications. Following CDS initiation, lead arrangers

(participants) are 1.15 (1.08) times more likely to initiate a new lending relationship with

borrowers having more extensive foreign operations relative to other borrowers.

Overall, the results reported in Table 4 suggest that once CDSs become available lead

27

arrangers and syndicate participants pursue new borrowers with high cross-selling potential,

consistent with their focus on commercial aspects of new lending relationships. Our findings also

imply that, while lead arrangers are likely to be the primary beneficiaries from cross-selling of

security underwriting, M&A advisory, and structured finance products, participants may also

enjoy some cross-selling opportunities, such as derivative and foreign operation transactions.

4.4 Lenders’ Monitoring Incentives

We next examine our prediction that the commercialization of the lending relationships leads

to lower monitoring incentives of non-relationship lenders in the post CDS-initiation period. We

perform our tests for the lead arranger-borrower relationships since lead arrangers perform the

primary monitoring of the borrower on behalf of syndicate participants.

4.4.1 Lead Arrangers’ Control Right

Because lenders retain stronger control rights when they aim to intensively monitor a

borrower, following prior literature, we measure lenders’ monitoring incentives by the strength of

their control rights (Roberts and Sufi 2009a, 2009b; Roberts 2015; Kim et al. 2018). Performance

(income statement-based) covenants act as tripwires and are used primarily to allocate control

rights to lenders when a borrower underperforms, while capital (balance sheet-based) covenants

address agency conflicts by aligning ex-ante incentives of a borrower and lenders (Christensen and

Nikolaev 2012; Christensen et al. 2016). Therefore, the higher (lower) proportion of performance

covenants should be associated with lenders’ stronger (weaker) control rights. We define

Performance Covenants as the ratio of the number of performance covenants to the sum of

performance and capital covenants in a loan contract.11

11 In line with Christensen and Nikolaev (2012), we classify cash interest coverage ratio, debt service coverage ratio, level of EBITDA, fixed charge coverage ratio, interest coverage ratio, ratio of debt to EBITDA, and ratio of senior debt to EBITDA covenants as performance covenants, while quick ratio, current ratio, debt to equity ratio, loan to

28

Roberts and Sufi (2009a) also suggest that performance pricing provisions affect lenders’

control rights, by influencing the allocation of bargaining power between a borrower and lenders

in ex post renegotiations. Performance pricing provisions define a pricing grid, which links a loan’s

interest rate to a borrower’s performance (Asquith et al. 2005), where the interest rate increases

(decreases) when a borrower’s performance deteriorates (improves). Because the interest rate

increasing provision typically imposes a sharp increase in the interest rate when performance

deteriorates, borrowers are incentivized to renegotiate the loan contract, thus allocating control

rights to lenders. We define Interest Increasing PP to be equal to 1 if a loan contract contains the

interest rate increasing performance pricing provisions and 0 otherwise. We measure Interest

Increasing PP based on the largest loan facility in the package.

We estimate the following model to examine the effect of CDS initiation on non-

relationship lead arrangers’ control rights.12

Control Rights = β 0 + β1POST + β2No Relationship-Lead Arranger

+ β3POST × No Relationship-Lead Arranger +β4Assets + β5ROA +

β6Loss + β7Leverage + β8Interest Coverage + β9Tangibility

+ β10Rated + β11Investment Grade + β12Amount + β13Maturity

+ β14Guarantor + β15PP + β16#Covenants + γFixed Effects + ε,

(3)

where Control Rights is either Performance Covenants or Interest Rate Increasing PP, as defined

above. No Relationship-Lead Arranger and all control variables are defined as previously. If non-

relationship lead arrangers retain weaker control rights post CDS initiation, we expect a negative

value ratio, ratio of debt to tangible net worth, leverage ratio, senior leverage ratio, and net worth requirement covenants as capital covenants. Financial covenants are identical to all loan facilities in the package. 12 If loan package has more than one lead arranger (6.5% of sample packages), all lead arrangers are accounted for in estimating model (3). Our results are unchanged when we exclude these deals from the analyses.

29

and significant coefficient on the interaction term POST × No Relationship-Lead Arranger. For

covenant specifications, we estimate Model (3) as Tobit (OLS) with firm and year (firm and year-

month) fixed effects. For performance pricing specifications, we estimate Model (3) as Logit

(OLS) with firm and year (firms and year-month) fixed effects.13

As we report in columns 1 and 2 of Panel A, Table 5, the coefficient on POST × No

Relationship-Lead Arranger is negative and significant in both specifications for Performance

Covenants, indicating that non-relationship lead arrangers impose a lower proportion of

performance covenants following CDS initiation relative to relationship lenders. Economically, in

the post-CDS initiation period, the proportion of performance covenants is 0.89 times lower when

non-relationship lead arrangers syndicate a loan relative to when a loan is syndicated by

relationship lead arrangers. We continue to find negative and significant coefficients on POST ×

No Relationship-Lead Arranger for Interest Rate Increasing PP specifications in column 3 and 4,

suggesting that CDS-protected non-relationship lead arrangers are less likely to impose provisions

that would increase the interest rate if a borrower underperforms. In the post-CDS initiation period,

non-relationship lead arrangers are 0.87 times less likely to impose the interest rate increasing

performance provisions relative to relationship lead arrangers. To support the robustness of these

findings, in columns 5 and 6 of Panel A, we perform the analyses for the weighted average Interest

Rate Increasing PP variable for all facilities in the deal, where weights are based on loan (facility)

size. We find very similar results.

Importantly, in contrast to the negative and significant coefficients on POST × No

Relationship-Lead Arranger, the coefficients on Post, which captures the effect of CDSs on control

rights of relationship lenders, are insignificant in all specifications. We infer that lead arrangers

13  We exclude #Covenants (PP) from the model when we examine performance covenants (interest increasing provisions), as it is highly correlated with the dependent variable.

30

that had a prior relationship with the borrower do not relinquish their control rights after CDS

initiation. This finding helps us understand prior evidence of the detrimental influence of CDSs on

lenders’ monitoring (e.g., Shan et al. 2015; Chakraborty et al. 2015). We show that weaker lender

monitoring is actually attributed to lenders that initiate a new relationship with the borrower

following CDS initiation, while relationship lenders do not reduce their monitoring efforts.

In untabulated analyses, we re-estimate Model (3) with covenant slack as the dependent

variable, as covenant tightness also determines lenders’ control rights (e.g., Dichev and Skinner

2002). Because financial ratios used in covenant definitions are adjusted extensively by lenders

(e.g., Dichev and Skinner 2002; Li 2016; Li et al. 2016), we focus on the debt-to-EBITDA

covenant, which slack can be measured reliably based on the DealScan data (e.g., Demerjian and

Owens 2016). This covenant is also the most frequently used for our sample loans. We measure

covenant tightness as the difference between a borrower’s debt-to-EBITDA ratio in the year

preceding the loan issuance measure minus the covenant threshold, divided by standard deviation

of the debt to EBITDA ratio over the previous 12 quarters (e.g., Dichev and Skinner 2002). We

find that, following CDS initiations, non-relationship lenders impose looser debt-to-EBITDA

covenants, as indicated by the positive and significant (albeit at the 10% level) coefficient on POST

× No Relationship-Lead Arranger. This evidence further suggests that lenders retain weaker

control rights when entering into new lending relationships following CDS initiation.

4.4.2 Lead Arrangers’ Share

Prior studies show that having more skin in the game (i.e., retaining a larger loan share)

incentivizes the lead arranger to monitor a borrower more intensively (e.g., Sufi 2007; Ball et al.

2008; Ivashina 2009). Therefore, we supplement the analyses of lenders’ monitoring incentives by

investigating whether non-relationship lead arrangers retain a lower loan share relative to

31

relationship lenders following CDS initiation.

Following prior studies (Ball et al. 2008; Ivashina 2009; Amiram et al. 2017), we perform

these analyses for the largest loan facility in the package and define Lead Arranger Share as the

proportion of the loan retained by the arranger. We re-estimate Model (3) with Lead Arranger

Share as the dependent variables and report the results in column 1 of Panel B. Consistent with

our predictions, we find negative and significant coefficients on POST × No Relationship-Lead

Arranger. Economically, in the post-CDS initiation period, lead arranger share is 0.92 times lower

when non-relationship lead arrangers syndicate a loan relative to when a loan is syndicated by

relationship lead arrangers. To support the robustness of these findings, in column 2 of Panel B,

we perform the analyses for the weighted average Lead Arranger Share for all facilities in the deal,

where weights are based on loan (facility) size. We find that our results are unchanged.

In line with our covenant- and performance-provision-based tests, the analyses presented in

Panel B also indicate that relationship and non-relationship lenders respond differently to CDS

initiation. Amiram et al.’s (2017) show that lead arrangers increase their loan share following CDS

initiation to compensate syndicate participants for the reduction in their monitoring incentives. We

find that this effect of CDSs is attributed primarily to lenders who had an established relationship

with the borrower before CDS initiation. In contrast to the negative and significant coefficient on

POST × No Relationship-Lead Arranger, the coefficient on Post, which reflects the effect of CDSs

on the loan share of relationship lenders, is positive and significant.

4.5 Non-relationship Lender Characteristics

To better understand the effect of CDS availability on non-relationship lenders’ monitoring

incentives, we next examine how the characteristics of non-relationship lenders change from the

pre-CDS initiation period to the post-CDS initiation period. If CDS-protected non-relationship

32

lead arrangers are less committed to intensively monitor the borrower, we predict that there is a

higher likelihood that lead arrangers with lower monitoring efficiency will initiate lending

relationships once CDSs on the borrower’s debt become available.

We examine this prediction using three measures of monitoring efficiency. First, we rely on

geographical distance between a lender and a borrower. Prior research suggests that greater

distance impedes the lender’s collection of private, soft information and increases its monitoring

costs, thus reducing monitoring efficiency (e.g., Petersen and Rajan 2002; Berger et al. 2005;

Degryse and Ongena, 2005; Sufi 2007; Agarwal and Hauswald, 2010; Dass and Massa 2011).

Therefore, given the reduced monitoring incentives of CDS-protected non-relationship lead

arrangers, we predict that, relative to relationship lead arrangers, they are likely to be more distant

from borrowers in the post- versus the pre-CDS initiation period. For lead arrangers headquartered

in the US, we measure the distance in miles between a lead arranger’s and a borrower’s

headquarters (Distance). As an additional measure of lender-borrower distance, we use an

indicator variable equal to 1 if the lender is not headquartered in US, and 0 otherwise (Foreign).

We next focus on the lead arranger’s reputation, which reflects its ability and commitment

to screen and monitor borrowers (e.g., Diamond 1989, 1991; Boot et al. 1993; Chemmanur and

Fulghieri 1994). Lender reputation is especially important in the syndicated loan market, where

repeated interactions occur between the arranger and the syndicate participants (Dennis and

Mullineaux 2000; Lee and Mullineaux 2004; Sufi 2007; Gopalan et al. 2011; Bushman and

Wittenberg-Moerman 2012). Following prior literature, we consider reputable arrangers as lenders

with a strong monitoring ability and incentives. Because CDS-protected non-relationship lead

arrangers are less likely to rigorously monitor the borrower, we predict that, relative to relationship

lead arrangers, non-relationship lead arrangers are less likely to be reputable in the post- versus

33

the pre-CDS initiation period. We define Reputation as an indicator variable equal to 1 if a lead

arranger is one of the top three lead arrangers, based on average market share in the syndicated

loan market over our sample period, and equals 0 otherwise (Bushman et al. 2010).14 We estimate

the following model to examine our predictions:

Lender Characteristics = β 0 + β1POST + β2No Relationship-Lead Arranger

+ β3POST × No Relationship-Lead Arranger + β4Assets

+ β5ROA + β6Loss + β7Leverage + β8Interest Coverage

+ β9Tangibility + β10Rated + β11Investment Grade + β12Amount

+ β13Maturity + β14Guarantor + γFixed Effects + ε,

(4)

where Lender Characteristics are Distance, Foreign and Reputation, as defined above. No

Relationship-Lead Arranger and all control variables are defined as previously. We expect a

positive (negative) and significant coefficient on the interaction term POST × No Relationship-

Lead Arranger in the Distance and Foreign (Reputation) specifications.

Consistent with our prediction, the coefficient on POST × No Relationship-Lead Arranger is

positive and significant for both lender-borrower distance measures. Economically, relative to

relationship lead arrangers, non-relationship lead arrangers are 143 miles further away from a

borrower and are 1.28 times more likely to be foreign in the post- versus the pre-CDS initiation

period. The coefficient on POST × No Relationship-Lead Arranger is negative and significant in

the Reputation specification. Relative to relationship lead arrangers, non-relationship lead

arrangers are 0.5 times less likely to be reputable during the post-CDS initiation period.

Overall, the results in Table 6 indicate that, following CDS initiation, non-relationship lead

arrangers are more likely to have lower monitoring efficiency, consistent with them being less

14 We measure market share based on Bloomberg’s lead-arranger league tables. Our findings are robust if we define reputable lead arrangers as the top five or top six rather than the top three.

34

likely to engage in costly monitoring of borrowers.

4.6 Resolving Agency Problems within the Loan Syndicate

Our findings presented in Tables 5 and 6 suggest a stronger adverse effect of CDSs on the

monitoring incentives of non-relationship relative to relationship lenders. A natural question arises

of how this weaker monitoring by CDS-protected non-relationship lead arrangers affects the

structure of loan syndicates. Syndicate participants typically maintain an arm’s-length relationship

with a borrower and delegate information collection, contractual negotiations, and borrower

screening and monitoring to the lead arranger. Participants thus face substantial agency issues

when they join a syndicate (e.g., Lee and Mullineaux 2004; Sufi 2007; Ivashina 2009). We expect

greater agency problems within the loan syndicate when it is arranged by a non-relationship

relative to a relationship lead arranger after CDS initiation. A stronger decline in non-relationship

lead arrangers’ monitoring incentives should cause syndicate participants to be concerned that

these arrangers will shirk more on their monitoring duties (e.g., Sufi 2007; Ivashina 2009). These

agency problems can be significantly mitigated if lenders with prior relationships with the

borrower become syndicate participants, because they have greater knowledge of the borrower’s

performance and creditworthiness (Sufi 2007). Thus, although we find that non-relationship

participants are more likely to join the syndicate after CDS initiation (Table 3), we predict that this

effect will be weaker when a loan is syndicated by a non-relationship lead arranger.

To test this prediction, we augment Model (1) with No Relationship-Lead Arranger and the

interaction term POST × No Relationship-Lead Arranger. If non-relationship participants are less

likely to join the syndicate in the post-CDS initiation period when the loan is syndicated by non-

relationship lead arrangers relative to when it is syndicated by relationship lead arrangers, we

expect a negative and significant coefficient on POST × No Relationship-Lead Arranger. We

35

report supporting evidence in Table 7. Non-relationship participants are 0.85 times less likely to

join syndicates arranged by non-relationship lead arrangers. Coupled with the positive and

significant coefficient on Post, as in previous analyses, these findings indicate that, while non-

relationship lenders are more likely to join the syndicate once CDSs become available, this effect

is attenuated when a loan is syndicated by a non-relationship lead arranger.15

Agency problems within the syndicate can also be mitigated if participants have a prior

relationship with the lead arranger (e.g., Sufi 2007; Bushman et al. 2017). In untabulated analyses,

we examine whether participants familiar with the lead arranger are more likely to join the

syndicate when the lead arranger initiates a lending relationship following CDS initiation. We find

no supporting evidence, but this result is generally consistent with Sufi (2007), who shows that,

when agency problems are severe, previous lead-arranger–participant relationships are much less

important than the previous relationships between the borrower and the participant.

4.7 Supplementary Analyses

In this section, we report a number of supplementary analyses. We begin with exploratory

tests of the effect of CDSs on the interest rate and size of loans syndicated by non-relationship lead

arrangers. On the one hand, these arrangers may bid aggressively for a loan by offering a low

spread or more credit. On the other hand, hedging loan exposure via CDSs is costly. If non-

relationship arrangers charge a low spread or have a high loan exposure, the new relationship may

not be profitable enough to pursue.

We re-estimate Model (3) with Interest Spread as the dependent variable and report the

results in column (1) and (2) of Panel A, Table 8. Interest Spread is the interest rate spread on the

15 Note that the sum of the coefficients on POST and POST × No Relationship – Lead Arranger is positive and significant, suggesting that, even when a loan is syndicated by a non-relationship lead arranger, the probability of non-relationship participants joining the syndicate is higher in the post- relative to the pre-CDS initiation period.

36

largest loan facility in the package. For robustness, we also perform the analyses for the weighted

average Interest Spread for all facilities in the deal, where weights are based on loan (facility) size.

The coefficient on the interaction term POST × No Relationship-Lead Arranger is insignificant in

the interest spread specifications, suggesting that non-relationship lenders do not charge a lower

spread relative to relationship lenders following CDS initiation.

In column (3) and (4) of Panel A, we next re-estimate Model (3) with Amount as the

dependent variables, where Amount is the natural logarithm of the loan amount of the largest

facility in the loan deal. We also perform analyses for the Deal Amount variable, measured by the

natural logarithm of loan deal (package) amount. Interestingly, the coefficient on POST × No

Relationship-Lead Arranger is negative and significant in both loan amount specifications.

Economically, in the post-CDS initiation period, the size of the largest loan (deal) size is 0.92

(0.93) times lower when a non-relationship lead arranger syndicates a loan relative to when it is

syndicated by a relationship lead arranger. This evidence indicates that CDS-protected non-

relationship lead arrangers have lower exposures to new borrowers, which allows them to reduce

hedging costs via CDSs. Given our findings of a substantial influence of cross-selling potential on

lead arrangers’ decisions to enter new relationships, we infer that, although these lenders have a

lower exposure to new borrowers to reduce hedging costs, borrowers still provide them with

opportunities to serve their additional financial needs.16

In the next set of analyses, to further illuminate lenders’ incentives to initiate new

relationships after the onset of CDS trading, we investigate how lenders’ performance affects the

relation between CDS trading and non-relationship lending. Lenders often cannot expand their

loan portfolio, because they are restricted by an internal credit risk management policy or face

16 We acknowledge that lower deal size may result in lower loan origination fees for lead arrangers, but expect cross-selling fess to be substantially higher than the decrease in origination fees.

37

regulatory capital constraints. CDS hedging offers lenders more flexible credit risk management

strategies and alleviate capital constraints (Streitz 2015; Shan et al. 2016). Therefore, we expect

that lenders experiencing low loan growth will be more likely to extend their loan portfolio and

seek new borrowers following CDS initiation. We also expect CDS initiation to more strongly

affect less profitable lenders. When hedging via CDSs and information revelation through CDS

spreads become available, these lenders are likely to be eager to pursue new relationships to

enhance their profitability. By expanding loan portfolios, they will benefit from both higher

interest income and fees from cross-selling additional services to new borrowers.

To test our predictions, we define an indicator variable Low Loan Growth to be equal to 1 if

the lender’s average loan growth over the three years before a loan’s issuance is below the sample

median and 0 otherwise. We augment Model (1) with Low Loan Growth and the interaction term

POST × Low Loan Growth. We present our findings in Panel B of Table 8 (these analyses are

restricted for lenders with data available from call reports). Consistent with our predictions, the

coefficient on POST × Low Loan Growth is positive and significant for both lead arranger and

participant subsamples. CDS-protected arrangers (participants) are 1.23 (1.29) times more likely

to initiate non-relationship lending when they experience lower loan growth.

We further define an indicator variable Low Profitability to be equal to 1 if the lender’s

average ROE over the three years before a loan’s issuance is below the sample median and 0

otherwise. We augment Model (1) with Low Profitability and the interaction term POST × Low

Profitability. As we report in Panel C of Table 8, the coefficient on POST × Low Profitability is

positive and significant for all specifications, in line with our prediction. Following CDS initiation,

lead arrangers (participants) are 1.52 (1.49) times more likely to issue loans to new borrowers

when they experience low profitability. Overall, the results suggest that lenders’ low loan growth

38

and profitability amplify the effect of CDSs on initiating lending relationships for both lead

arrangers and syndicate participants.

5. Conclusion

We examine how the development of the CDS market affects lenders’ incentives to initiate

new lending relationships. Due to the information advantage of incumbent relationship lenders,

non-relationship lenders face adverse selection problems when competing for a loan deal, which

inhibits their relationships with new borrowers (Rajan 1992). We predict that an opportunity to

hedge against credit losses and the revelation of private information through CDS spreads

significantly mitigate non-relationship lenders’ adverse selection concerns, increasing their

willingness to compete for loans to new borrowers. Consistent with our prediction, we find that

both lead arrangers and syndicate participants are more likely to initiate new relationships once

CDSs become available. Our findings also highlight the commercialization of the resulting new

lending relationships. CDS-protected non-relationship lenders enhance the profitability of their

new relationships by pursuing new borrowers with high cross-selling potential and by reducing

their own monitoring efforts. Further emphasizing the lower commitment of non-relationship

lenders to monitoring the borrower once CDSs become available, we find that, relative to

relationship lenders, non-relationship lenders are likely to be more distant from borrowers, foreign,

and less reputable in the post-CDS initiation period. We further show that the increase in the

likelihood of non-relationship participants joining a syndicate following CDS initiation is

attenuated if the loan is syndicated by a non-relationship lead arranger, which mitigates agency

problems within the syndicate exacerbated by the lower monitoring incentives of non-relationship

lead arrangers.

Our paper contributes to the growing literature on the influence of the CDS market on private

39

lending by exploring how CDSs affect the formation of new lending relationships and the

commercialization of these relationships. We further contribute to the expansive literature on

relationship lending by showing that CDS availability increases the willingness of non-relationship

lenders to compete for new borrowers and thus enhances competition in the loan market. By

documenting that a borrower’s cross-selling potential significantly affects lenders’ decisions to

enter new relationships, we also add to the growing literature on the importance of noncredit

business opportunities in private lending.

40

References

Acharya, V., and T. Johnson (2007). Insider trading in credit derivatives. Journal of Financial Economics, 84 (1), 110–141.

Agarwal, S., and R. Hauswald (2010). Distance and private information in lending. The Review of Financial Studies, 23(7), 2757-2788.

Amiram, D., W. Beaver, W. Landsman, and J. Zhao (2017). The effects of cds trading initiation on the structure of syndicated loans. Journal of Financial Economics, 126 (2), 364-382.

Ashcraft, A., and J. Santos (2009). Has the cds market lowered the cost of corporate debt? Journal of Monetary Economics, 56 (4), 514–523.

Asquith, P., A. Beatty, and J. Weber (2005). Performance pricing in bank debt contracts. Journal of Accounting and Economics, 40 (1), 101-128.

Ball, R., R. Bushman, and F. Vasvari (2008). The debt contracting value of accounting information and loan syndicate structure. Journal of Accounting Research, 46 (2), 247-287.

Batta, G., J. Qiu, and F. Yu (2016). Credit derivatives and analyst behavior. The Accounting Review, 91 (5), 1315–1343.

Berger, A., N. Miller, M. Petersen, R. Rajan, and J. Stein (2005). Does function follow organizational form? Evidence from the lending practices of large and small banks. Journal of Financial Economics, 76 (2), 237-269.

Berger, A., and G. Udell (1995). Relationship lending and lines of credit in small firm finance. Journal of Business, 68 (3), 351–381.

Berlin, M., and L. Mester (1992). Debt covenants and renegotiation. Journal of Financial Intermediation, 2 (2), 95–133.

Bertrand, M., and S. Mullainathan (2003). Enjoying the quiet life? Corporate governance and managerial preferences. Journal of Political Economy, 111 (5), 1043-1075.

Bharath, S., S. Dahiya, A. Saunders, and A. Srinivasan (2009). Lending relationships and loan contract terms. The Review of Financial Studies, 24 (4), 1141–1203.

Bolton, P., and M. Oehmke (2011). Credit default swaps and the empty creditor problem. Review of Financial Studies, 24 (8), 2617–2655.

Boot, A., S. Greenbaum, and A. Thakor. (1993). Reputation and discretion in financial contracting. The American economic review, 83 (5), 1165-1183.

Boot, A. (2000). Relationship banking: What do we know? Journal of Financial Intermediation, 9 (1), 7-25.

Bradley, M., and M. Roberts (2015). The structure and pricing of corporate debt covenants. The Quarterly Journal of Finance, 5 (02), 1550001.

Buis, M.L. (2010). Stata tip 87: Interpretation of interactions in non-linear models. The Stata Journal, 10 (2), 305-308.

Bushman, R., C. Williams, and R. Wittenberg Moerman (2017). The informational role of the media in private lending. Journal of Accounting Research 55 (1), 115-152.

Bushman, R., A. Smith, and R. Wittenberg-Moerman, 2010, Price discovery and dissemination of private information by loan syndicate participants, Journal of Accounting Research 48, 921-972.

41

Bushman, R. and R. Wittenberg Moerman (2012). The role of bank reputation in “certifying” future performance implications of borrowers’ accounting numbers. Journal of Accounting Research, 50 (4), 883-930.

Chakraborty, I., S. Chava, and R. Ganduri (2015). Credit default swaps and moral hazard in bank lending, Working paper, Georgia Institute of Technology.

Champagne, C., and L. Kryzanowski (2007). Are current syndicated loan alliances related to past alliances? Journal of Banking and Finance, 31 (10), 3145-3161.

Chemmanur, T., and P. Fulghieri, (1994). Investment bank reputation, information production, and financial intermediation. The Journal of Finance, 49 (1), 57-79.

Christensen, H., and V. Nikolaev (2012). Capital versus performance covenants in debt contracts. Journal of Accounting Research, 50 (1), 75–116.

Christensen, H., V. Nikolaev, and R. Wittenberg Moerman (2016). Accounting information in financial contracting: The incomplete contract theory perspective. Journal of Accounting Research, 54 (2), 397–435.

Dass, N., and M. Massa (2011). The impact of a strong bank-firm relationship on the borrowing firm. The Review of Financial Studies, 24(4), 1204-1260.

Dennis, S., and D. Mullineaux (2000). Syndicated loans. Journal of Financial Intermediation, 9 (4), 404-426.

Degryse, H., and S. Ongena (2005). Distance, lending relationships, and competition. The Journal of Finance, 60 (1), 231-266.

Demerjian, P., and E. Owens (2016). Measuring the probability of financial covenant violation in private debt contracts. Journal of Accounting and Economics, 61 (2), 433-447.

Diamond, D. W. (1989). Reputation acquisition in debt markets. Journal of Political Economy, 97(4), 828-862.

Diamond, D. (1991). Monitoring and reputation: The choice between bank loans and directly placed debt. Journal of Political Economy, 99 (4), 689–721.

Dichev, I., and D. Skinner (2002). Large–sample evidence on the debt covenant hypothesis. Journal of Accounting Research, 40 (4), 1091-1123.

Drucker, S., and M. Puri (2005). On the benefits of concurrent lending and underwriting. The Journal of Finance, 60 (6), 2763–2799.

Euromoney (2014). Why no amount of cross-selling can make corporate lending profitable. February 26.

Gopalan, R., V. Nanda and V. Yerramilli (2011). Does poor performance damage the reputation of financial intermediaries? Evidence from the loan syndication market. The Journal of Finance, 66 (6), 2083-2120.

Greene, W. (2004). The behaviour of the maximum likelihood estimator of limited dependent variable models in the presence of fixed effects. The Econometrics Journal, 7 (1), 98-119.

Gntay, L., and D. Hackbarth (2010). Corporate bond credit spreads and forecast dispersion. Journal of Banking & Finance, 34 (10), 2328–2345.

Fang, L., V. Ivashina and J. Lerner (2013). Combining banking with private equity investing. The Review of Financial Studies, 26 (9), 2139-2173.

Financial Times, 2005, Banks warned about insider trading in credit derivatives, April 25.

42

Heider, F., and A. Ljungqvist (2015). As certain as debt and taxes: Estimating the tax sensitivity of leverage from state tax changes. Journal of Financial Economics 118 (3), 684–712.

Holmstrom, B., and J. Tirole (1997). Financial intermediation, loanable funds, and the real sector. The Quarterly Journal of Economics, 112 (3), 663–691.

Hu, H., and B. Black (2008). Debt, equity and hybrid decoupling: Governance and systemic risk implications. European Financial Management, 14 (4), 663– 709.

ISDA (2013). Summaries of market survey results, Available at: http://www.isda.org/statistics/recent.html

Ivashina, V. (2009). Asymmetric information effects on loan spreads. Journal of Financial Economics, 92 (2), 300–319.

Ivashina, V., and A. Kovner (2011). The private equity advantage: Leveraged buyout firms and relationship banking. The Review of Financial Studies, 24 (7), 2462–2498.

Ivashina, V., and Z. Sun (2011). Institutional stock trading on loan market information. Journal of Financial Economics, 100 (2), 284-303.

Kim, J., P. Shroff, D. Vyas, and R. Wittenberg Moerman (2018). Credit default swaps and managers’ voluntary disclosure, Journal of Accounting Research, 56, 953-988.

Lee, S., and D. Mullineaux (2004). Monitoring, financial distress, and the structure of commercial lending syndicates. Financial Management, 33 (3), 107-130.

Liaw, K. (2011). The business of investment banking: a comprehensive overview, New Jersey: Wiley, 2011. pp. 1-7.

Li, N. (2016). Performance measures in earnings‐based financial covenants in debt contracts. Journal of Accounting Research, 54 (5), 1149-1186.

Li, N., F. Vasvari and R. Wittenberg-Moerman (2016). Dynamic threshold values in earnings-based covenants. Journal of Accounting and Economics, 61 (2), 605-6

Maddala, G. (1987). Limited dependent variable models using panel data. Journal of Human Resources, 22 (3), 307-338.

Mansi, S., W. Maxwell, and D. Miller (2011). Analyst forecast characteristics and the cost of debt. Review of Accounting Studies, 16 (1), 116–142.

Martin, X. and S. Roychowdhury (2015). Do financial market developments influence accounting practices? Credit default swaps and borrowers reporting conservatism. Journal of Accounting and Economics, 59 (1), 80–104.

Minton, B., R. Stulz, and R. Williamson (2009). How much do banks use credit derivatives to hedge loans? Journal of Financial Services Research, 35 (1), 1–31.

Morrison, A. (2005). Credit derivatives, disintermediation, and investment decisions. The Journal of Business, 78 (2), 621–648.

Neilson, J., K. Wang, C. Williams and B. Xie, (2017). Offsetable derivative exposures and financial stability. Working Paper. University of Michigan.

Oliver Wyman (2006). Get smart or get out: structured finance, Available at: http://www.oliverwyman.com/.

Parlour, C., and A. Winton (2013). Laying off credit risk: Loan sales versus credit default swaps. Journal of Financial Economics, 107 (1), 25–45.

43

Petersen, M., and R. Rajan (1994). The benefits of lending relationships: Evidence from small business data. The Journal of Finance, 49 (1), 3–37.

Petersen, M., and R. Rajan (1995). The effect of credit market competition on lending relationships. The Quarterly Journal of Economics, 110 (2), 407–443.

Petersen, M., and R. Rajan (2002). Does distance still matter? The information revolution in small business lending. The Journal of Finance, 57(6), 2533-2570.

Qiu, J., and F. Yu (2012). Endogenous liquidity in credit derivatives. Journal of Financial Economics, 103 (3), 611-631.

Rajan, R. (1992). Insiders and outsiders: The choice between informed and arm’s length debt. The Journal of Finance, 47 (4), 1367–1400.

Roberts, M., and A. Sufi (2009a). Renegotiation of financial contracts: Evidence from private credit agreements. Journal of Financial Economics, 93 (2), 159–184.

Roberts, M., and A. Sufi (2009b). Control rights and capital structure: An empirical investigation. The Journal of Finance, 64 (4), 1657–1695.

Roberts, M. (2015). The role of dynamic renegotiation and asymmetric information in financial contracting. Journal of Financial Economics, 116 (1), 61-81.

Sapienza, P. (2002). The effects of banking mergers on loan contracts. The Journal of Finance, 57, 329-367.

Saretto, A., and H. Tookes (2013). Corporate leverage, debt maturity, and credit supply: The role of credit default swaps. Review of Financial Studies, 26 (5), 1190–1247.

Shan, S., D. Tang, and A. Winton (2015). Do credit derivatives lower the value of creditor control rights? Evidence from debt covenants. Working paper, University of Minnesota.

Shan, S., D. Tang, and H. Yan (2016). Regulation-induced financial innovation: The case of credit default swaps and bank capital. Working paper, Shanghai University of Finance and Economics.

Shan, S., D. Tang, and H. Yan (2016). How does CDS trading affect bank lending relationships, Working paper, Shanghai University of Finance and Economics.

Sharpe, S. (1990). Asymmetric information, bank lending, and implicit contracts: A stylized model of customer relationships. The Journal of Finance, 45 (4), 1069–1087.

Standard&Poor’s (2011). A guide to the loan market. Standard&Poor’s New York. Streitz, D. (2015). The impact of credit default swap trading on loan syndication. Review of

Finance, 20 (1), 265–286. Subrahmanyam, M., D. Tang, and S. Wang (2014). Does the tail wag the dog?: The effect of credit

default swaps on credit risk. The Review of Financial Studies, 27 (10), 2927–2960. Sufi, A. (2007). Information asymmetry and financing arrangements: Evidence from syndicated

loans. The Journal of Finance, 62 (2), 629–668. Valta, P. (2012). Competition and the cost of debt. Journal of Financial Economics, 105, 661-682. Yasuda, A. (2005). Do bank relationships affect the firm’s underwriter choice in the corporate

bond underwriting market? The Journal of Finance, 60 (3), 1259– 1292. Zhao, J. (2017). The economics of bank cross-firm selling: The value of borrower board

connectedness and opacity, Working paper, The University of North Carolina at Chapel Hill.

44

APPENDIX A Variable Definitions

Variable Definition

Amount = The natural logarithm of loan amounts of the largest facility in the loan package (DealScan).

Assets = The natural logarithm of total assets, measured in the year preceding a loan’s issuance (Compustat).

CDS Initiation = An indicator variable equal to 1 for the year of CDS initiation, and 0 otherwise (Markit).

Deal Amount = The natural logarithm of the package amount (DealScan).

Distance = The distance in miles between the lead arranger and the borrower (Compustat).

Forex = The ratio of notional amounts of FX derivatives to the lenders’ total assets, measured in the year preceding a loan’s issuance (Call Report).

Foreign = An indicator variable equal to 1 if a lead arranger is not headquartered in USand 0 otherwise (DealScan).

,Guarantor = An indicator variable equal to 1 if the loan is guaranteed, and 0 otherwise (DealScan).

High-CS Potential-Derivatives = An indicator variable equal to 1 if the ratio of derivative-related word count to total word count in the 10-K filing in the year preceding a loan’s issuance is greater than the sample median, and 0 otherwise (10-K).

High-CS Potential-Foreign = An indicator variable equal to 1 if the ratio of the number of a borrower’s foreign operating segments to its total number of operating segments in the year preceding a loan’s issuance is greater than the sample median, and 0 otherwise (Compustat).

High-CS Potential-Industry = An indicator variable equal to 1 if a borrower is in the aviation, shipbuilding, oil, refinery, gas, or telecommunication industries, and 0 otherwise (Compustat).

High-CS Potential-M&A = An indicator variable equal to 1 if the borrower engages in M&A deals within 3 years after a loan’s issuance, and 0 otherwise (SDC).

High-CS Potential-Media = An indicator variable equal to 1 if a media covers topics related to cross-selling opportunities (e.g., public offering, merger, acquisition, and investment) within 3 years before a loan’s issuance, and 0 otherwise (RavenPack).

Interest Coverage = The ratio of earnings before interest and taxes to interest expense, measured in the year preceding a loan’s issuance (Compustat).

Interest Spread = The natural logarithm of the all-in-drawn spread of the largest facility in the package (DealScan).

Interest Spread (weighted average)

= The natural logarithm of the weighted average all-in-drawn spread for all facilities in the deal, where weights are based on loan (facility) size (DealScan).

Interest Rate Increase PP = An indicator variable equal to 1 if a loan contains an interest rate increasing performance-pricing provision, and 0 otherwise (DealScan).

Investment Grade = An indicator variable equal to 1 if the firm has an investment-grade rating in the year prior to a loan’s issuance, and 0 otherwise (Compustat, Mergent FISD).

45

APPENDIX A (continued)

Variable Definition

Last Loan = An indicator variable equal to 1 for the last loan issued prior to a borrower's CDS initiation date, and 0 otherwise (DealScan, Markit).

Lead Arranger Share = The proportion (%) of the loan retained by the lead arranger, measured for the largest facility in the loan package (DealScan).

Lead Arranger Share (weighted average)

= The weighted average lead arranger share variable for all facilities in the deal, where weights are based on loan (facility) size (DealScan).

Leverage = The ratio of total liabilities to total assets, measured in the year preceding a loan’s issuance (Compustat).

Loss = An indicator variable equal to 1 if a borrower’s net income in the year preceding a loan’s issuance is less than 0, and 0 otherwise (Compustat).

Low Coverage = An indicator variable equal to 1 if the number of equity analysts following the borrower in the year of a loan’s issuance is less than the sample median, and 0 otherwise (I/B/E/S).

Low Loan Growth = An indicator variable equal to 1 if the lender's average loan growth over the 3-year period prior to a loan’s issuance is below the sample median, and 0 otherwise (Call Report).

Low ROE = An indicator variable equal to 1 if the lender's average ROE over the 3-year period prior to the loan issuance is below the sample median, and 0 otherwise (Call Report).

Maturity = The maturity of a loan in months (DealScan). MTB = The market value divided by the book value of equity, measured in the year

prior to the year under consideration (Compustat).

No Relationship - Lead Arranger = For lead arrangers that syndicate loans to CDS firms, an indicator variable equal to 1 if the lead arranger has not syndicated a borrower’s loans prior to the CDS initiation date, and 0 otherwise. For lead arrangers that syndicate loans to non-CDS firms, an indicator variable equal to 1 if the lead arranger has not syndicated a borrower’s loans prior to the issuance date of the loan under consideration, and 0 otherwise (DealScan, Markit).

No Relationship - Participant = For syndicate participants in loans to CDS firms, an indicator variable equal to 1 if the participant has not participated in a borrower’s loans prior to the CDS initiation date, and 0 otherwise. For syndicate participants in loans to non-CDS firms, an indicator variable equal to 1 if the participant has notparticipated in a borrower’s loans prior to the issuance date of the loan under consideration, and 0 otherwise (DealScan, Markit).

Performance Covenants = The ratio of the number of performance covenants divided by the sum of performance and capital covenants in a loan contract (DealScan).

POST = An indicator variable equal to 1 if the loan is issued after the CDS trading initiation date, and 0 otherwise (Markit).

PP = An indicator variable equal to 1 if the loan has a performance pricing provision, and 0 otherwise (DealScan).

Profit Margin = The ratio of net income to sales, measured in the year prior to the year under consideration (Compustat).

46

APPENDIX A (continued)

Variable Definition

Rated = An indicator variable equal to 1 if the firm is rated by S&P or Moody's in the year prior to a loan’s issuance, and 0 otherwise (Compustat, Mergent FISD).

Reputation = An indicator variable equal to 1 if a lead arranger is one of the top 3 lead arrangers by the average market share over the sample period), and 0 otherwise (DealScan, Bloomberg).

Return Volatility = The standard deviation of the firm’s monthly stock return, measured in the year prior to the year under consideration (Compustat, CRSP).

ROA = The ratio of net income to total assets (Compustat).

Tangibility = The ratio of property, plant, and equipment to total assets (Compustat).

#Covenants = The total number of financial covenants (DealScan).

47

Appendix B Additional Analyses

TABLE B1

Borrower Information Opacity

This table examines whether the effect of CDS trading on non-relationship lending is more pronounced for more opaque borrowers. Columns 1 and 2 (3 and 4) report the analysis for the lead arranger (participant) specifications. Columns 1 and 3 (2 and 4) present results using a logit (OLS) model. We include firm and year (year-month) fixed effects in the logit (OLS) model. t-statistics in parentheses are based on standard errors clustered at the firm level. ***, **, and * indicate significance at the 0.01, 0.05, and 0.10 levels, respectively. All variables are defined in Appendix A.

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

POST 0.534*** 0.092*** 0.499*** 0.112***

(3.86) (3.57) (6.84) (7.09)

Low Coverage -0.179** -0.035** -0.055 -0.012

(-2.43) (-2.33) (-0.88) (-0.94)

POST*Low Coverage 0.338** 0.079*** 0.092 0.018

(2.30) (2.82) (1.16) (1.06)

Model Logit OLS Logit OLS

Controls Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

Year-month FE No Yes No Yes

Observations 24,582 27,196 175,677 177,815

Adj. (Pseudo) 0.029 0.181 0.015 0.145

48

TABLE B2 Propensity Score Matching

This table presents the propensity score matching (PSM) analyses. Panel A provides results of the first-stage probit model of CDS initiation. Panel B reports the difference in the means of the explanatory variables between the CDS firms and matched non-CDS firms to provide evidence of covariate balancing in the PSM estimation. t-statistics in parentheses are based on standard errors clustered at the firm level. ***, **, and * indicate significance at the 0.01, 0.05, and 0.10 levels, respectively. All variables are defined in Appendix A. Panel A: First-stage CDS Initiation Probit Model CDS Initiation

Rated 0.848***

(13.46)

Investment Grade 0.362***

(6.39)

Assets 0.107***

(5.44)

Leverage -0.019

(-0.20)

Profit Margin -0.002***

(-3.03)

Return Volatility 0.075

(1.14)

MTB 0.004

(0.96)

Model Probit

Observations 27,553

Adj. (Pseudo) 0.192

49

TABLE B2 (continued) Propensity Score Matching

Panel B: Covariate Balancing

Means Difference in means

CDS firms non-CDS firms (t-stats)

Rated 0.910 0.901 0.009

(0.54)

Investment Grade 0.565 0.587 (0.022)

(-0.85)

Assets 8.407 8.283 0.124*

(1.77)

Leverage 0.647 0.651 (0.004)

(-0.41)

Profit Margin (1.446) (0.003) (1.443)

(-0.98)

Return Volatility 0.043 0.041 0.002

(0.14)

MTB 3.020 3.090 (0.070)

(-0.30)

50

TABLE B3 Industries with High Cross-selling Potential

This table presents SIC codes of industries we classify as having high cross-selling potential.

Industry SIC code SIC industry name

Oil, Refinery and Gas

1311 CRUDE PETROLEUM & NATURAL GAS

1381 DRILLING OIL & GAS WELLS

1382 OIL & GAS FIELD EXPLORATION SERVICES

1389 OIL & GAS FIELD SERVICES, NEC

2911 PETROLEUM REFINING

4922 NATURAL GAS TRANSMISSION

4923 NATURAL GAS TRANSMISISON & DISTRIBUTION

4924 NATURAL GAS DISTRIBUTION

Aviation

3720 AIRCRAFT & PARTS

3721 AIRCRAFT

3724 AIRCRAFT ENGINES & ENGINE PARTS

3728 AIRCRAFT PARTS & AUXILIARY EQUIPMENT, NEC

4512 AIR TRANSPORTATION, SCHEDULED

4522 AIR TRANSPORTATION, NONSCHEDULED

Telecommunication 4813 TELEPHONE COMMUNICATIONS

Shipbuilding 3730 SHIP & BOAT BUILDING & REPAIRING

51

TABLE 1 Descriptive Statistics

This table provides descriptive statistics for the lead arranger and participant samples. All variables are defined in Appendix A.

Lead arranger sample Participant sample

N Mean Median SD N Mean Median SD

No Relationship 27,457 0.38 0.00 0.49 177,936 0.44 0.00 0.50

Assets 27,457 7.79 7.70 2.06 177,936 8.28 8.20 1.69

ROA 27,457 0.03 0.04 0.09 177,936 0.03 0.04 0.07

Loss 27,457 0.21 0.00 0.41 177,936 0.16 0.00 0.37

Leverage 27,457 0.64 0.63 0.23 177,936 0.66 0.65 0.21

Interest Coverage 27,457 11.02 3.55 26.81 177,936 9.65 3.79 22.36

Tangibility 27,457 0.60 0.56 0.40 177,936 0.59 0.55 0.40

Rated 27,457 0.58 0.00 0.49 177,936 0.72 1.00 0.45

Investment Grade 27,457 0.33 1.00 0.47 177,936 0.42 1.00 0.49

Amount 27,457 19.47 0.00 1.54 177,936 20.06 0.00 1.14

Maturity 27,457 46.69 19.52 25.65 177,936 48.33 20.03 23.40

Guarantor 27,457 0.10 55.00 0.30 177,936 0.11 60.00 0.31

PP 27,457 0.42 1.00 0.49 177,936 0.57 1.00 0.50

#Covenants 27,457 1.14 0.00 1.31 177,936 1.38 0.00 1.32

52

TABLE 2 The Effect of CDS Trading on Non-relationship Lending

This table examines the effect of CDSs on non-relationship lending. Panel A reports results of the primary tests. Panel B reports results of the analysis that investigates whether the parallel trend assumption holds for our tests. Panel C reports results for the short-window analysis, in which observations of CDS firms are limited to 3 years before and after CDS initiation. Columns 1 and 2 (3 and 4) report the analysis for the lead arranger (participant) specifications. Columns 1 and 3 (2 and 4) present results using a logit (OLS) model. We include firm and year (year-month) fixed effects in the logit (OLS) model. t-statistics in parentheses are based on standard errors clustered at the firm level. ***, **, and * indicate significance at the 0.01, 0.05, and 0.10 levels, respectively. All variables are defined in Appendix A. Panel A: Primary Tests

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

POST 0.778*** 0.149*** 0.563*** 0.125***

(8.12) (8.46) (10.79) (11.39)

Assets 0.011 0.003 -0.180*** -0.037***

(0.22) (0.31) (-4.98) (-4.94)

ROA -0.053 -0.002 0.049 0.007

(-0.16) (-0.03) (0.17) (0.11)

Loss 0.077 0.021 0.016 0.002

(1.19) (1.58) (0.31) (0.21)

Leverage -0.263* -0.054* -0.206* -0.039

(-1.81) (-1.87) (-1.70) (-1.50)

Interest Coverage 0.000 0.000 0.001* 0.000**

(-0.24) (-0.37) (1.86) (2.00)

Tangibility 0.505*** 0.103*** 0.019 0.009

(3.02) (3.18) (0.17) (0.38)

Rated -0.014 0.000 0.032 0.006

(-0.18) (0.01) (0.53) (0.47)

Investment Grade 0.023 0.007 -0.128** -0.027**

(0.24) (0.39) (-2.37) (-2.37)

Amount -0.157*** -0.033*** 0.041* 0.008*

(-5.42) (-5.92) (1.81) (1.69)

Maturity 0.008*** 0.002*** 0.010*** 0.002***

(8.10) (8.80) (14.12) (14.40)

Guarantor 0.230*** 0.047*** 0.243*** 0.052***

(3.29) (3.40) (5.12) (5.19)

Model Logit OLS Logit OLS

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

Year-month FE No Yes No Yes

Observations 24,582 27,196 175,677 177,815

Adj. (Pseudo) 0.029 0.181 0.015 0.145

53

TABLE 2 (continued) The Effect of CDS Trading on Non-relationship Lending

Panel B: Pre-trend Analyses

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

Last Loan -0.151 -0.029 -0.031 -0.002

(-1.18) (-1.33) (-0.47) (-0.13)

POST 0.727*** 0.138*** 0.551*** 0.124***

(7.06) (7.20) (9.40) (9.95)

Model Logit OLS Logit OLS

Controls Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

Year-month FE No Yes No Yes

Observations 24,582 27,196 175,677 177,815

Adj. (Pseudo) 0.029 0.181 0.015 0.145

Panel C: Short-window Analyses

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

POST 0.354*** 0.073*** 0.249*** 0.050***

(3.18) (3.63) (4.18) (4.18)

Model Logit OLS Logit OLS

Controls Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

Year-month FE No Yes No Yes

Observations 18,377 21,423 129,685 131,879

Adj. (Pseudo) 0.025 0.145 0.018 0.161

54

TABLE 3 Addressing Endogeneity Concerns

This table presents results of the analysis to address endogeneity concerns. Panel A reports results based on propensity score matching (PSM) analysis using the full sample. Panel B reports results of PSM analysis using a short-window sample, which limit observations to 3 years before and after the CDS initiation date (the pseudo-CDS initiation date for non-CDS firms). In Panels A and Panel B, Columns 1 and 2 (3 and 4) report the analysis for the lead arranger (participant) specifications. Columns 1 and 3 (2 and 4) present results using a logit (OLS) model. We include firm and year (year-month) fixed effects in the logit (OLS) model. Panel C presents results of the second-stage instrumental variable (IV) regression analyses, estimated simultaneously with the first-stage CDS initiation model. Column 1 (2) reports the IV regression analysis for the lead arranger (participant) specification. We include firm and year fixed effects in the model. Across all models, t-statistics in parentheses are based on standard errors clustered at the firm level. ***, **, and * indicate significance at the 0.01, 0.05, and 0.10 levels, respectively. All variables are defined in Appendix A. Panel A: PSM Analyses—Full Sample

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

POST -0.102 -0.014 -0.221*** -0.042***

(-0.85) (-0.72) (-3.27 ) (-3.30)

POST*CDS Traded 0.367** 0.074*** 0.266*** 0.060***

(2.13) (2.64) (2.83) (3.36)

Model Logit OLS Logit OLS

Controls Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

Year-month FE No Yes No Yes

Observations 16,366 17,284 143,972 144,533

Adj. (Pseudo) 0.100 0.295 0.130 0.172

Panel B: PSM Analyses—Short Window

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

POST -0.069 -0.014 -0.205*** -0.034***

(-0.47) (-0.62) (-3.00) (-2.66)

POST*CDS Traded 0.335* 0.064** 0.203** 0.046**

(1.66) (1.99) (2.07) (2.38)

Model Logit OLS Logit OLS

Controls Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

Year-month FE No Yes No Yes

Observations 6,476 7,875 68,340 68,954

Adj. (Pseudo) 0.040 0.275 0.011 0.158

55

Panel C: Instrumental Variable (IV) Analyses

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2)

POST 1.209** 10.114***

(2.11) (4.19)

Assets -0.040 -0.393**

(-1.62) (-2.15)

ROA -0.025 -0.366

(-0.17) (-0.41)

Loss 0.000 0.032

(-0.01) (0.20)

Leverage -0.052 -1.276**

(-0.65) (-2.39)

Interest Coverage 0.000 0.006**

(0.63) (2.42)

Tangibility 0.223* 1.420**

(1.77) (2.43)

Rated -0.023 0.132

(-0.51) (0.56)

Investment Grade -0.059 -0.497*

(-1.31) (-1.90)

Amount -0.022** 0.034

(-2.09) (0.46)

Maturity 0.001 -0.006*

(1.44) (-1.91)

Guarantor 0.031 0.095

(1.25) (0.69)

Model OLS OLS

Firm FE Yes Yes

Year FE Yes Yes

Observations Yes Yes

Adj. (Pseudo) 8,483 33,759

56

TABLE 4 Borrower Cross-selling Potential

This table examines whether the effect of CDS trading on non-relationship lending is more pronounced when a borrower has high cross-selling potential. Panels A, B, C, D, and E report results of the analyses in which cross-selling potential is measured based on media coverage of public offerings, mergers, acquisitions, and investment topics in borrower-specific articles preceding a loan’s issuance (High CS Potential-Media), a borrower’s M&A transactions following a loan’s issuance (High CS Potential-M&A), a borrower’s participation in the aviation, shipbuilding, oil, refinery, gas, or telecommunication industries (High CS Potential-Industry), a borrower’s demand for derivative-related products (High CS Potential-Derivatives), and the extent of a borrower’s foreign operations (High CS Potential-Foreign), respectively. In all panels, Columns 1 and 2 (3 and 4) report the analyses for the lead arranger (participant) specifications. Columns 1 and 3 (2 and 4) present results using a logit (OLS) model. We include firm and year (year-month) fixed effects in the logit (OLS) model. t-statistics in parentheses are based on standard errors clustered at the firm level. ***, **, and * indicate significance at the 0.01, 0.05, and 0.10 levels, respectively. All other variables are defined in Appendix A. Panel A: Media Coverage

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

POST 0.737*** 0.135*** 0.398*** 0.088***

(5.71) (5.83) (6.15) (6.54)

High CS Potential-Media -0.066 -0.010 0.089* 0.016*

(-0.95) (-0.79) (1.99) (1.74)

POST*High CS Potential-Media 0.250** 0.047** -0.071 -0.010

(2.03) (2.12) (-1.18) (-0.81)

Model Logit OLS Logit OLS

Controls Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

Year-month FE No Yes No Yes

Observations 19,025 22,148 141,898 143,912

Adj. (Pseudo) 0.027 0.219 0.012 0.151

57

TABLE 4 (continued) Borrower Cross-selling Potential

Panel B: Mergers and Acquisitions

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

POST 0.756*** 0.145*** 0.572*** 0.127***

(7.84) (8.16) (10.94) (11.52)

High CS Potential-M&A -0.152 -0.025 0.118 0.025

(-1.38) (-1.31) (1.56) (1.64)

POST* High CS Potential-M&A 0.489** 0.087** -0.142 -0.034

(2.13) (2.10) (-1.03) (-1.13)

Model Logit OLS Logit OLS

Controls Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

Year-month FE No Yes No Yes

Observations 24,582 27,196 175,677 177,815

Adj. (Pseudo) 0.029 0.181 0.015 0.145

Panel C: Industry Type

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

POST 0.706*** 0.136*** 0.538*** 0.119***

(7.10) (7.34) (9.83) (10.42)

POST*High CS Potential-Industry 0.520** 0.089** 0.173 0.042

(2.11) (2.03) (1.37) (1.51)

Model Logit OLS Logit OLS

Controls Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

Year-month FE No Yes No Yes

Observations 24,582 27,196 175,677 177,815

Adj. (Pseudo) 0.029 0.181 0.015 0.145

58

TABLE 4 (continued) Borrower Cross-selling Potential

Panel D: Derivative Transactions

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

POST 0.597*** 0.115*** 0.395*** 0.089***

(4.81) (4.94) (5.73) (6.02)

High CS Potential-Derivatives -0.027 -0.006 -0.014 -0.004

(-0.45) (-0.50) (-0.34) (-0.43)

POST*High CS Potential-Derivatives 0.298** 0.055** 0.253*** 0.054***

(2.38) (2.31) (3.78) (3.75)

Model Logit OLS Logit OLS

Controls Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

Year-month FE No Yes No Yes

Observations 24,582 27,196 175,677 177,815

Adj. (Pseudo) 0.029 0.181 0.015 0.145

Panel E: Foreign Operations

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

POST 0.667*** 0.128*** 0.512*** 0.114***

(6.21) (6.29) (9.17) (9.55)

High CS Potential-Foreign -0.107 -0.025* -0.045 -0.009

(-1.46) (-1.77) (-0.91) (-0.88)

POST*High CS Potential-Foreign 0.295** 0.056** 0.153** 0.034**

(2.07) (2.12) (2.01) (2.09)

Model Logit OLS Logit OLS

Controls Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

Year-month FE No Yes No Yes

Observations 24,582 27,196 175,677 177,815

Adj. (Pseudo) 0.029 0.181 0.015 0.145

59

TABLE 5

Non-relationship Lead Arrangers’ Monitoring Incentives

This table presents results of the analyses that examine non-relationship lead arrangers’ monitoring incentives following CDS initiation. In Panel A, we measure monitoring incentives by the strength of lead arranger’s control rights: in Columns (1) and (2) we examine the proportion of performance covenants in the total of performance and capital covenants (Performance Covenants) and, in Columns (3) – (6), the existence of the interest-rate-increasing pricing provision (Interest Rate Increasing PP). In Panel B, we measure monitoring incentives by the share retained by the lead arranger (Lead Arranger Share). We include firm and year (year-month) fixed effects in logit and Tobit (OLS) models. t-statistics in parentheses are based on standard errors clustered at the firm level. ***, **, and * indicate significance at the 0.01, 0.05, and 0.10 levels, respectively. All variables are defined in Appendix A.

60

TABLE 5 (continued) Non-relationship Lead Arrangers’ Monitoring Incentives

Panel A: Lead Arrangers' Control Rights

Performance Covenants Interest Rate Increasing PP

Interest Rate Increasing PP (weighted average)

(1) (2) (3) (4) (5) (6)

POST 0.081* -0.009 0.091 0.015 0.050 0.013

(1.88) (-0.65) (0.82) (0.89) (0.45) (0.81)

No Relationship -0.004 -0.003 0.008 0.000 0.052 0.002 - Lead Arranger (-0.16) (-0.49) (0.15) (-0.07) (0.99) (0.33)

POST*No Relationship -0.263*** -0.041*** -0.251** -0.039** -0.238** -0.039** - Lead Arranger (-4.98) (-2.98) (-2.16) (-2.37) (-2.06) (-2.42)

Assets -0.280*** -0.014* -0.135** -0.015* -0.114* -0.014

(-19.76) (-1.94) (-2.13) (-1.72) (-1.80) (-1.64)

ROA -0.077 -0.028 -0.767 -0.055 -0.581 -0.047

(-0.46) (-0.53) (-1.53) (-0.82) (-1.15) (-0.71)

Loss 0.163*** 0.026** -0.258*** -0.030** -0.251*** -0.031***

(4.67) (2.30) (-2.93) (-2.45) (-2.88) (-2.59)

Leverage 0.361*** 0.034 -1.502*** -0.173*** -1.474*** -0.172***

(6.13) (1.38) (-7.01) (-6.37) (-6.98) (-6.42)

Interest Coverage 0.002*** 0.000* -0.001 0.000 -0.001 0.000

(5.17) (1.88) (-0.80) (-0.75) (-0.56) (-0.62)

Tangibility -0.351*** -0.007 0.030 -0.003 0.102 0.001

(-8.88) (-0.29) (0.16) (-0.10) (0.54) (0.04)

Rated 0.221*** 0.008 0.092 0.020 0.044 0.018

(6.48) (0.66) (0.84) (1.28) (0.41) (1.21)

Investment Grade -0.438*** -0.031** 0.415*** 0.072*** 0.403*** 0.069***

(-10.80) (-2.04) (3.99) (4.09) (3.88) (4.02)

Amount 0.115*** 0.022*** 0.448*** 0.056*** 0.495*** 0.056***

(8.76) (5.54) (11.36) (12.05) (12.17) (12.38)

Maturity 0.004*** 0.001*** 0.002 0.000 0.002 0.000

(8.15) (4.37) (0.90) (1.63) (1.13) (1.37)

Guarantor 0.681*** 0.158*** 0.382*** 0.056*** 0.414*** 0.056***

(19.47) (11.64) (4.35) (4.01) (4.74) (4.13)

PP 1.388*** 0.310***

(54.22) (36.74)

#Covenants 0.740*** 0.109*** 0.768*** 0.109***

(27.11) (30.32) (28.44) (31.00)

Model Tobit OLS Logit OLS Logit OLS

Firm FE Yes Yes Yes Yes Yes Yes

Year FE Yes No Yes No Yes No

Year-month FE No Yes No Yes No Yes

Observations 27,457 27,196 20,819 27,196 21,159 27,196

Adj. (Pseudo) N/A 0.517 0.164 0.298 0.180 0.307

61

TABLE 5 (continued) Non-relationship Lead Arranger’ Monitoring Incentives

Panel B: Lead Arrangers’ “Skin in the Game”

Lead Arranger Share

Lead Arranger Share (weighted average)

(1) (2)

POST 3.449*** 5.018***

(3.63) (5.26)

No Relationship 3.784*** 2.871*** - Lead Arranger (6.59) (4.51)

POST*No Relationship -3.334*** -3.532*** - Lead Arranger (-2.96) (-3.29)

Assets -2.103*** -1.882**

(-3.12) (-2.58)

ROA 1.359 0.880

(0.22) (0.13)

Loss 3.246*** 2.552**

(3.47) (2.49)

Leverage 0.169 1.122

(0.06) (0.40)

Interest Coverage 0.024* 0.033**

(1.72) (2.06)

Tangibility -1.374 -1.880

(-0.64) (-0.86)

Rated 0.884 2.337*

(0.71) (1.86)

Investment Grade 0.862 1.977*

(0.77) (1.79)

Amount -10.094*** -9.967***

(-19.08) (-16.10)

Maturity -0.098*** -0.096***

(-6.20) (-5.76)

Guarantor 1.128 0.760

(1.28) (0.84)

PP -6.405*** -5.318***

(-8.29) (-6.35)

#Covenants -0.230 -0.841**

(-0.67) (-2.22)

Model OLS OLS

Firm FE Yes Yes

Year FE No No

Year-month FE Yes Yes

Observations 8,398 8,398

Adj. (Pseudo) 0.718 0.659

62

TABLE 6 Non-relationship Lead Arrangers’ Characteristics

This table presents results of the analyses that examine non-relationship lead arranger characteristics following CDS initiation. In Columns (1) and (2), we examine whether the non-relationship lead arranger is more likely to be foreign (Foreign). In Column (3), we examine the distance (Distance) between the borrower and the lead arranger. In Columns (4) and (5), we examine whether the non-relationship lead arranger is more likely to be reputable (Reputation) We include firm and year (year-month) fixed effects in the logit and OLS models. t-statistics in parentheses are based on standard errors clustered at the firm level. ***, **, and * indicate significance at the 0.01, 0.05, and 0.10 levels, respectively. All variables are defined in Appendix A.

Foreign Distance Reputation

(1) (2) (3) (4) (5)

POST -0.160 -0.026** -59.583*** 0.011 0.008

(-1.08) (-2.06) (-3.51) (0.09) (0.52)

No Relationship - Lead Arranger

0.814*** (11.23)

0.075*** (10.92)

12.172 (1.09)

-0.317***

(-5.35) -0.034***

(-4.93) POST*No Relationship - Lead Arranger

0.388** (2.00)

0.069*** (3.21)

143.067***

(4.34)

-0.346* (-1.78)

-0.051** (-2.05)

Assets 0.264*** 0.033*** 10.839 0.190*** 0.015**

(3.86) (5.05) (0.95) (2.87) (2.08)

ROA 0.603 0.062 -91.424 0.305 0.031

(1.30) (1.61) (-1.07) (0.72) (0.61)

Loss -0.072 -0.009 8.443 0.004 0.000

(-0.72) (-0.97) (0.57) (0.05) (0.01)

Leverage 0.157 0.019 31.960 -0.250 -0.029

(0.75) (1.04) (0.90) (-1.12) (-1.16)

Interest Coverage -0.001 0.000 -0.168 0.000 0.000

(-0.75) (-1.21) (-0.81) (0.20) (-0.13)

Tangibility -0.345* -0.022 -45.302 0.350 0.023

(-1.65) (-1.14) (-1.31) (1.59) (0.96)

Rated -0.018 -0.001 -30.878 0.160 0.016

(-0.14) (-0.13) (-1.57) (1.41) (1.25)

Investment Grade -0.185 -0.016 3.469 0.010 0.002

(-1.43) (-1.26) (0.19) (0.08) (0.15)

Amount -0.192*** -0.024*** -7.003 0.046 0.006

(-4.29) (-4.83) (-1.05) (1.17) (1.39)

Maturity 0.010*** 0.001*** 0.493** -0.005*** -0.000***

(6.75) (5.87) (2.57) (-3.68) (-3.22)

Guarantor 0.257** 0.021* -12.489 0.037 0.003

(2.11) (1.85) (-0.67) (0.41) (0.30)

Model Logit OLS OLS Logit OLS

Firm FE Yes Yes Yes Yes Yes

Year FE Yes No No Yes No

Year-Month FE No Yes Yes No Yes

Observations 13,541 27,196 16,222 17,056 27,196

Adj. (Pseudo) 0.062 0.505 0.735 0.043 0.405

63

TABLE 7 Agency Problems within the Syndicate

This table presents results of the analyses that investigate how agency problems within the syndicate are mitigated following CDS initiation. Column 1 (2) presents results using a logit (OLS) model. We include firm and year (year-month) fixed effects in the logit (OLS) model. t-statistics in parentheses are based on standard errors clustered at the firm level. ***, **, and * indicate significance at the 0.01, 0.05, and 0.10 levels, respectively. All other variables are defined in Appendix A.

No Relationship - Participant

(1) (2)

POST 0.559*** 0.120***

(10.12) (10.46)

No Relationship - Lead Arranger 0.745*** 0.161***

(21.24) (21.42)

POST*No Relationship - Lead Arranger -0.343*** -0.068***

(-5.57) (-5.07)

Model Logit OLS

Controls Yes Yes

Firm FE Yes Yes

Year FE Yes No

Year-month FE No Yes

Observations 175,677 177,815

Adj. (Pseudo) 0.025 0.154

64

TABLE 8 Supplementary Analyses

This table presents results of the supplementary analyses. Panel A examines the effect of CDSs on the interest spread and size of loans syndicated by non-relationship lead arrangers. In Columns (1) and (2), we examine the interest spread (Interest Spread) and in Columns (3) and (4), we examine the size of a loan (Amount, Deal Amount). Analyses include firm and year-month fixed effects. Panels B and C examine whether the effect of CDS trading on non-relationship lending is more pronounced for lenders with low loan growth (Low Loan Growth) and low profitability (Low Profitability), respectively. Columns 1 and 2 (3 and 4) report the analyses for the lead arranger (participant) specifications. Columns 1 and 3 (2 and 4) present results using a logit (OLS) model. We include firm and year (year-month) fixed effects in the logit (OLS) model. t-statistics in parentheses are based on standard errors clustered at the firm level. ***, **, and * indicate significance at the 0.01, 0.05, and 0.10 levels, respectively. All variables are defined in Appendix 1. Panel A: Interest Spread and Loan Amount

Loan Spread

Loan Spread (weighted average)

Amount Deal Amount

(1) (2) (3) (4)

POST 0.040 0.038 -0.030 -0.043

(1.55) (1.49) (-0.92) (-1.26)

No Relationship 0.059*** 0.058*** -0.058*** -0.038** - Lead Arranger (5.81) (5.75) (-3.84) (-2.41)

POST*No Relationship -0.020 -0.019 -0.084** -0.077* - Lead Arranger (-0.81) (-0.79) (-2.19) (-1.95)

Assets -0.111*** -0.112*** 0.462*** 0.479***

(-8.72) (-8.76) (18.47) (18.95)

ROA -0.468*** -0.454*** 0.408** 0.441***

(-4.96) (-4.97) (2.49) (2.59)

Loss 0.164*** 0.164*** -0.077*** -0.070**

(8.02) (8.03) (-2.66) (-2.36)

Leverage 0.313*** 0.316*** 0.020 0.081

(8.48) (8.50) (0.28) (1.10)

Interest Coverage -0.001*** -0.001*** 0.001*** 0.001***

(-6.83) (-6.96) (2.61) (2.89)

Tangibility -0.309*** -0.311*** -0.015 -0.003

(-6.89) (-7.09) (-0.20) (-0.03)

Rated 0.040* 0.037* -0.100*** -0.109***

(1.77) (1.67) (-3.61) (-3.75)

Investment Grade -0.203*** -0.194*** 0.029 -0.001

(-6.00) (-6.20) (0.76) (-0.03)

Amount -0.060*** -0.057***

(-6.46) (-6.16)

Maturity 0.000 0.000 0.004*** 0.005***

(1.19) (1.28) (6.90) (8.73)

Guarantor 0.013 0.017 -0.066** -0.072**

(0.79) (0.99) (-2.36) (-2.40)

65

Table 8 Panel A (continued) PP -0.102*** -0.102*** 0.214*** 0.197***

(-7.10) (-7.65) (11.40) (10.38)

#Covenants 0.055*** 0.055*** 0.031*** 0.059***

(11.09) (11.25) (4.34) (7.75)

Model OLS OLS OLS OLS

Firm FE Yes Yes Yes Yes

Year FE No No No No

Year-month FE Yes Yes Yes Yes

Observations 27,196 27,196 27,196 27,196

Adj. (Pseudo) ) 0.756 0.76 0.790 0.774

Panel B: Lender Loan Growth

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

POST 0.181 0.043 0.374*** 0.066***

(1.00) (1.61) (4.31) (4.53)

Low Loan Growth -0.408*** -0.062*** -0.362*** -0.062***

(-4.32) (-4.45) (-9.71) (-10.08)

POST*Low Loan Growth 0.382** 0.059** 0.464*** 0.080***

(2.03) (2.23) (6.79) (6.67)

Model Logit OLS Logit OLS

Controls Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

Year-month FE No Yes No Yes

Observations 6,149 9,333 34,909 36,589

Adj. (Pseudo) 0.043 0.382 0.035 0.215

Panel C: Lender Profitability

No Relationship - Lead Arranger

No Relationship - Participant

(1) (2) (3) (4)

POST -0.139 0.001 0.170* 0.040***

(-0.61) (0.04) (1.80) (2.78)

Low Profitability 0.038 0.015 0.043 0.008

(0.41) (0.98) (1.14) (1.20)

POST*Low Profitability 0.912*** 0.134*** 0.812*** 0.136***

(3.57) (3.66) (9.42) (9.52)

Model Logit OLS Logit OLS

Controls Yes Yes Yes Yes

Firm FE Yes Yes Yes Yes

Year FE Yes No Yes No

66

Year-month FE No Yes No Yes

Observations 6,149 9,333 34,909 36,589

Adj. (Pseudo) 0.044 0.176 0.038 0.150